O WhatsApp está avaliando (na versão beta 26.37.10.15) a implementação do suporte ao acesso de até três contas simultâneas no mesmo dispositivo.
A opção de adicionar uma terceira conta é bem semelhante ao que já foi implementado pela Meta ao adicionar o suporte a uma segunda conta no mensageiro, exibindo a foto de perfil e o número de telefone para facilitar a identificação na hora de alternar entre os perfis.
Outro recurso que a Meta avalia implementar (na versão beta 26.37.10.16) é a possibilidade de sincronizar para todos os participantes da conversa determinadas informações, como o papel de parede (ou os temas oferecidos pelo próprio WhatsApp) importado a partir da biblioteca de fotos do aparelho, além da cor dos balões de conversa.
A definição dessa configuração para todos os participantes da conversa permite que todos os usuários visualizem as interações de um jeito semelhante, tanto para conversas individuais do WhatsApp quanto para conversas em grupo (os usuários são alertados caso qualquer alteração nesse sentido seja feita).
A sincronização desses elementos também conta com a criptografia de ponta a ponta oferecida em outros recursos do aplicativo.
No momento, ambos os recursos estão sendo testados pelo WhatsApp e estão indisponíveis inclusive para quem utiliza a versão beta do app via TestFlight.
O Gemini Live permite que os usuários do aplicativo do Gemini para iPhones e iPads tenham uma conversa contínua e mais natural com a inteligência artificial — podendo inclusive interrompê-la quando precisar, bastando começar a falar enquanto ela responde em voz alta.
Porém, você pode simplesmente desativar esse comportamento, impedindo que o Gemini seja interrompido quando você fala algo em voz alta.
O Mac se tornou um importante aliado de quem usa a máquina para rodar modelos de inteligência artificial locais — não é à toa que os Macs mini estão sendo muito visados nesse período de boom no mercado.
Caso use o Gemini (um dos principais nomes de chatbots, ao lado do ChatGPT e do Claude), é possível habilitar um ajuste que mantém o computador ativo enquanto o app do Gemini estiver aberto, impedindo que ele entre em repouso.
iOS 27 includes a small new Phone app detail for birthdays: a fireworks animation can appear when you call someone whose birthday is saved in Contacts. To see the reported effect, a contact needs a birthday entered in its card and the call needs to take place on that date. The demonstration does not establish […]
On iOS 27, your iPhone can turn a voicemail into a to-do. With Apple Intelligence, the Phone app shows what Apple calls “Siri suggested actions” based on the voicemail transcript, such as adding a reminder or creating a calendar event. Apple’s iOS 27 guide documents the feature and its off switch, Suggestions in Voicemail, at […]
Fato é que alguns outros apps também receberam atualizações com essa mesma proposta — além, é claro, de incluir outras novidades. Vamos conferi-los a seguir.
Tripsy
Além de suporte ao iOS 27, o Tripsy — que apareceu na keynote da Apple, diga-se — foi atualizado com algumas outras novidades e melhorias que estão disponíveis como parte dos recursos do Tripsy Pro, entre elas:
Smart Import: importe atividades, locais e reservas de qualquer URL ou documento usando o analisador baseado em IA.
Smart Insights: utilizando os Apple Foundation Models, o aplicativo analisa e compreende o conteúdo de documentos, anexos e emails relacionados às atividades, extraindo apenas as informações relevantes e úteis para apresentá-las aos usuários.
Falando de melhorias, agora é possível encontrar todos os hotéis da sua viagem nas sugestões de localização e dispensar sugestões de partida de aeroportos de que não precisa mais. Ademais, foram aprimoradas as estatísticas de viagem, a estabilidade do mapa e os layouts de importação no Mac.
Por fim, também foi adicionado suporte ao widget extragrande no iPhone para uma melhor visualização do itinerário.
Loop
O player de podcasts Loop ganhou, na sua atualização mais recente, suporte à Siri e aos Atalhos, bem como ao recurso gestos de cabeça dos AirPods compatíveis para pular capítulos de episódios mesmo com o app em segundo plano.
Personalidades do Loop: nas configurações (em “Sobre mim”), agora é possível definir as personalidades que o Loop usará ao gerar o texto da seção “Para você”.
Métricas e estatísticas de escuta: três anéis para você acompanhar seus objetivos de escuta diários e mensais, acompanhando quantos episódios foram finalizados no mês, tempo de escuta diário e variedade de podcasts escutados.
Novos widgets: agora é possível ver o texto da seção “Para você” na Tela de Início em dois tamanhos de widgets (pequeno e médio) além de também ser possível ver seus anéis das métricas de escuta nas Telas de Início e Bloqueada.
O que ouvir em seguida: seção onde novos episódios dos podcasts que você marcou como favoritos e fixados são mostrados para fácil acesso.
O app também recebeu diversas melhorias em áreas como reprodutor de podcasts, busca e episódios baixados, além de uma melhor performance com o app em segundo plano, consumindo muito menos processamento e bateria.
FitWoody
Mais do que o suporte à Siri AI, que possibilita registrar dados por voz com ajuda da assistente, o app de saúde FitWoody agora também permite encontrar informações facilmente com o Spotlight e conta com suporte ampliado ao Handoff. O app também ganhou uma versão nativa para o macOS.
Divulgação/FitWoody
Quanto às melhorias, agora o FitWoody abre diretamente na própria tela, bem como sua pontuação do sono responde a cada noite à medida que ela acontece, de modo que a tela inicial já está completa logo pela manhã.
Steady
O aplicativo de treino Steady chegou à versão 3.0, que tem como destaque o recurso Importação Mágica — de modo que uma foto da ficha da academia, um PDF ou uma mensagem vira um programa de treino editável. Esse recurso usa os novos modelos de IA da Apple em iPhones compatíveis com Apple Intelligence e iOS 27, sem exigir o Steady Pro.
Além disso, o app ganhou planejamento de programas mais inteligente, navegador de programas, recuperação do treino, entre outros recursos.
Divulgação/Steady
A atualização também traz drop sets e deixa o app do Apple Watch mais completo, com lista de exercícios, registro de esforço pela Digital Crown e resumo do treino, reduzindo a necessidade de pegar o iPhone enquanto ele permanece por perto.
Ulysses
Já o popular aplicativo de escrita Ulysses ganhou, graças às capacidades da Siri AI, reconhecimento de contexto na tela e um botão dedicado na barra de ferramentas do macOS que abre a nova interface da assistente para escrever, reescrever, revisar e realizar outras tarefas com texto.
Ainda com relação à barra de ferramentas do aplicativo, agora é possível que os usuários a personalizem totalmente com botões de navegação, blocos, espaços, sugestões e muito mais.
We just released a very nice update across all platforms. The new version is ready for OS 27, features support for Siri AI, and introduces a fully customizable toolbar on macOS.
The update is available now. You can find the release notes on our website.https://t.co/TO7PFbVjGt
A atualização também inclui novos ícones de grupo com pictogramas e aprimora o suporte à acessibilidade “ao editar condições de filtro e ícones de grupo”. Por fim, o app recebeu ajustes visuais para melhor alinhar ao visual do iOS 27 e do macOS Golden Gate 27.
Aplicativos que utilizam uma premissa simples e a transformam em uma experiência interessante são, particularmente, meus favoritos. Exemplo disso é o Farwave, um app para iPhone e iPad desenvolvido pelo brasileiro Fernando Bunn que permite encontrar estações de rádio de todo o mundo girando um globo 3D até uma determinada região e tocando em uma estação.
A interação baseia-se na localização geográfica, e não em pesquisas por texto. Ou seja, os usuários giram o globo, selecionam uma região e tocam em uma estação, tudo pela mesma interface.
Divulgação/Farwave
Há, ainda, um botão chamado “Take me anywhere” (“Leve-me a qualquer lugar”, em tradução direta) que seleciona uma opção aleatória para ouvintes que não querem procurar um local específico; além disso, o aplicativo salva o histórico de reprodução para que as pessoas possam retornar a uma estação mais tarde.
Além do rádio, o app também organiza podcasts e músicas utilizando a mesma metáfora do globo. Diferentemente das estações de rádio, assinantes Pro podem pesquisar programas ou adicionar manualmente um feed RSS. O aplicativo também oferece suporte a filas de reprodução, velocidade de reprodução ajustável e downloads para ouvir offline.
Assinantes do plano Pro também podem importar seus próprios arquivos de áudio de um computador ou do aplicativo Arquivos. Cada artista aparece como um “sinal” próprio, e as faixas importadas permanecem no dispositivo para reprodução offline.
Siri AI is beginning its public rollout as Apple transforms its familiar assistant into a substantially more conversational system built around personal context, onscreen awareness, web knowledge, and the ability to perform actions across apps.
The beta is available first in English on compatible devices running iOS 27, iPadOS 27, macOS 27, watchOS 27, and visionOS 27. French, Japanese, Korean, Portuguese, and Spanish support is scheduled for October. Availability remains more limited in some regions, with Siri AI initially unavailable on iPhone, iPad, and Apple Watch in the European Union and unavailable in China while Apple addresses regulatory requirements.
The release follows several years of gradual Apple Intelligence expansion, but Siri AI represents a more fundamental change. Instead of handling each request largely as an isolated command, the assistant can maintain conversations, find relevant personal information, understand what is displayed on a device, and combine those capabilities to complete multistep tasks.
Image Credit: Apple Inc.
Siri AI can connect personal context with app actions
The system can search across a user’s messages, email, photos, notes, and other information when responding to a request. Apple offers an example in which someone asks what a family member suggested doing during an upcoming visit. Siri can locate the relevant conversation, find a recipe previously sent by email, and then place the ingredients into a Reminders grocery list.
That combination of retrieval and action is central to the new design. Siri AI can draft an email, edit and share photos, create calendar events, manage reminders, control media, and perform other supported actions without requiring users to manually move information between applications.
Third-party apps can participate through Apple’s App Intents architecture. Existing integrations include tasks such as messaging through WhatsApp or playing an audiobook in Audible, with Apple also previewing deeper capabilities in applications including Microsoft Outlook, Notability, and Tripsy.
Onscreen awareness adds another layer. Siri can identify information currently visible on a device and act on it. Someone viewing a sports website, for example, can ask Siri to identify upcoming home games and add them to Calendar without manually copying dates.
Apple has also expanded Siri beyond voice. On iPhone, users can invoke the assistant with the side button, voice command, or by swiping down from Dynamic Island. On iPad and Mac, Siri AI is integrated into Spotlight, while systemwide context menus can send selected files, images, or text directly to the assistant.
Image Credit: Apple Inc.
Siri AI gets its own app and conversation history
Siri AI is no longer confined to a temporary interface that disappears once a request is complete.
Apple has introduced a dedicated Siri app where users can begin new conversations or revisit previous ones. Conversation history can sync privately through iCloud across iPhone, iPad, Mac, Apple Watch, and Apple Vision Pro, allowing a discussion started on one device to continue on another.
Follow-up questions can retain the context of an earlier answer, giving Siri a more conventional conversational interface for longer exchanges.
The assistant also has access to broad web knowledge for questions that cannot be answered through personal information or local device data. Apple says Siri can retrieve current information for topics ranging from sports results to travel planning and nearby businesses.
Visual Intelligence expands alongside the assistant. On iPhone, a dedicated Siri mode inside Camera can identify and respond to objects or information in front of the user. A concert poster, for example, can become the starting point for questions about performance times followed by the creation of Calendar events.
On iPad, Visual Intelligence is integrated with screenshots. Mac users can select content on their displays through a keyboard shortcut, while Vision Pro users can ask Siri about app windows or physical objects simply by looking at them.
Writing Tools are similarly integrated across the system. Siri can generate drafts, revise existing text from natural-language instructions, and provide writing suggestions. In Mail and Messages, Apple says it can take a user’s typical communication style with a particular recipient into account, including tone and punctuation patterns.
Image Credit: Apple Inc.
Siri AI runs across on-device and private cloud models
Siri AI is powered by a new generation of Apple Foundation Models developed in collaboration with Google using Gemini technology, marking a notable shift in Apple’s AI infrastructure.
The architecture divides processing between models running directly on the device and larger models operating through Private Cloud Compute. Apple says requests sent to Private Cloud Compute are processed without personal data being stored or made accessible to the company.
Core resources such as the Spotlight index and Apple’s App Toolbox can remain on-device, allowing Siri to retrieve local information and perform supported actions without sending every request to a remote server.
More capable devices can also use AFM Core Advanced, Apple’s highest-performance on-device model. It powers improvements including more expressive Siri voices and advanced systemwide dictation that automatically handles capitalization, punctuation, and formatting.
Compatibility varies by feature. Apple Intelligence supports iPhone 16 models and later, iPhone 15 Pro and Pro Max, M1 or later iPads and Macs, iPad mini with A17 Pro, Apple Vision Pro, and supported Apple Watch models when paired with a nearby Apple Intelligence-enabled iPhone.
AFM Core Advanced requires newer hardware, including iPhone Duo, the iPhone 18 Pro line, iPhone Air, iPhone 17 Pro models, qualifying M4 or later iPads with at least 12GB of unified memory, M3 or later Macs with at least 12GB, and Apple Vision Pro with M5.
Image Credit: Apple Inc.
Siri AI arrives with limits as Apple expands access
The initial Siri AI release remains explicitly labeled beta, and some of the surrounding Apple Intelligence features are scheduled for later updates.
Apple also confirms that features relying on server-side models can be subject to daily usage limits. That includes Siri AI, cloud-based photo editing, Image Playground, and AFM 3 Cloud models used through Shortcuts. Limits can vary depending on the feature, request complexity, system demand, and other factors.
Apple says expanded access to server-dependent features will eventually be available for a fee, introducing a future paid layer to parts of Apple Intelligence even as the core system remains integrated into supported devices.
Apple will also expand its intelligence capabilities on Apple Watch later in 2026. Planned Audio Intelligence features on Apple Watch Series 12 and Ultra 4 include Live Rewind, which can display text representing the previous 15 seconds of a conversation, and Siri Recap, which generates high-level summaries that can be reviewed later in the Siri app.
Those functions are designed around a hardware-isolated Secure Exclave on the S11 chip, with Apple saying raw audio is processed in isolation and immediately deleted rather than stored as recordings. Live Rewind and Siri Recap are scheduled to enter beta later in 2026, making the current Siri AI launch the beginning of a staged expansion rather than the final form of Apple’s new assistant.
A new WhatsApp Business MCP server lets developers use AI coding agents like Claude, Cursor, Codex, and ChatGPT to handle setup, messaging templates, testing, and troubleshooting.
Nesta terça-feira, confira, aproveite e economize com a nossa seleção de promoções na App Store!
Planet of Lana é uma bela aventura, desenvolvida pelo Wishfully Studios e publicada nas plataformas móveis pela Playdigious, e nosso destaque do dia.
O jogo nos conquista primeiro pelos olhos: cada cenário parece uma pintura em movimento, com cores vibrantes, que transforma o planeta alienígena em algo ao mesmo tempo acolhedor e misterioso.
A jornada de Lana e seu companheiro Mui se destaca pela atmosfera contemplativa, criando uma experiência com um tom esperançoso. O design dos puzzles é intuitivo e nos incentiva a observar o ambiente e experimentar soluções criativas.
Confira um vídeo:
Um jogo cuja história passeia pelo equilíbrio, natureza e conexão. Uma oferta imperdível! 🥹
Abaixo outros jogos/apps que, juntos, somam quase R$130 em descontos:
Jogos
Jogo de aventura.
Jogo de cartas.
Diversão matemática.
Aplicativos
Calculadora.
Utilitário para códigos QR.
Fotos panorâmicas.
Aproveite as ofertas e até amanhã! Lembrando que elas são sempre por tempo limitado, então é bom correr. ✈️
O fim de ano já está batendo à porta, o que significa que logo veremos nossas redes sociais serem inundadas por retrospectivas de serviços de streaming como o Spotify, que anunciou hoje uma novidade que deverá agradar pais e responsáveis com crianças pequenas em casa.
Mais precisamente, a plataforma sueca começou a liberar uma nova opção nos seus ajustes que, quando ativada, impede que músicas infantis, cantigas de ninar e outras canções tradicionalmente associadas aos pequenos apareçam no Wrapped, a famosa retrospectiva de final de ano do Spotify.
Disponível tanto para usuários gratuitos quanto para assinantes do Premium, a novidade afeta o algoritmo de recomendação do serviço como um todo, o que significa que essas músicas também deixarão de aparecer em playlists como “Descobertas da Semana” e “Daily Mix”, só para citar alguns exemplos.
Ainda segundo as informações, esse ajuste também funciona retroativamente, o que significa que músicas infantis reproduzidas antes da sua ativação também serão ignoradas pelo serviço de streaming.
Fazendo isso, você elimina rapidamente a influência de músicas infantis e familiares nas suas recomendações, mantendo o foco no seu gosto musical. Assim, você pode tocar músicas infantis quando seus filhos estiverem por perto, como em uma viagem de carro ou na hora de dormir, sem deixar que essas músicas influenciem o que você ouvirá em seguida.
Para crianças e adolescentes que desejam conferir suas próprias retrospectivas, o Spotify oferece as chamadas Managed accounts, as quais foram expandidas recentemente e dão aos pais/responsáveis controle sobre a experiência de seus filhos na plataforma. Disponíveis em 16 países (incluindo Brasil e Portugal), essas contas podem ser configuradas por usuários com qualquer tipo de assinatura.
Para ativar o ajuste supracitado, basta abrir o app do Spotify, tocar na sua foto de perfil e selecionar “Configurações e Privacidade”. Feito isso, vá até “Conteúdo e tela” e, depois, desative o ajuste “Include kids and family music in taste profile” (algo como “Inclua músicas infantis e familiares no perfil de gosto”).
Divulgação/Spotify
De acordo com o Spotify, as alterações podem levar até 48 horas para serem refletidas em suas recomendações. Também é possível que a novidade demore um pouco para aparecer para todos.
Last year Ruxandra Teslo, a policy analyst who focuses on clinical trials, posted an idea for supercharging medical AI systems: Use data from failed biotech companies.
By bidding at their bankruptcy proceedings, she proposed, it might be possible to obtain detailed regulatory filings, manufacturing strategies, and safety data—types of information usually considered trade secrets. She called these documents “biotech’s lost archive” and said they could be used to help train AIs that would act as powerful copilots in the often opaque drug approval process.
Today the OpenAI Foundation, the nonprofit parent of OpenAI, said it would fund her idea as part of a new effort it calls Public Data for Health, which aims to help artificial intelligence make big leaps in medicine by paying to create “high-quality scientific datasets.”
The basic idea is that AI isn’t going to be capable of making important breakthroughs in curing disease unless researchers can feed the models much more information than they have so far.
“Everyone is recognizing that data is the biggest bottleneck in successfully applying AI to biology,” says Morgan Levine, a former vice president for computation at Altos Labs, a longevity company.
In its initial round of data grants, the OpenAI Foundation also announced that it would give $40 million to a program to collect data about novel cancer vaccines at the University of North Carolina, Chapel Hill, and support OpenAdmet, a group that runs competitions in which researchers try to predict drug effects.
Teslo’s idea for a biotech archive received $500,000 and will be pursued by 1Day Sooner, an advocacy group representing clinical trial volunteers, which she advises.
“We expect many remaining breakthroughs in preventing and curing disease to come from pairing the intelligence of new models with more observations of the world—in other words, more data,” the OpenAI Foundation said in a statement.
OpenAI started as a nonprofit, but leader Sam Altman restructured it to form a for-profit corporation that develops new models, launches products, and is now planning an initial public offering of stock that could value it at $1 trillion.
Because the foundation holds a 26% equity stake in OpenAI, it is now be on track to become the richest charitable organization on the planet, potentially sitting on $250 billion in stock value. (By comparison, the Gates Foundation and a trust associated with it held about $180 billion at the end of 2025.)
Making good use of that kind of money will not be easy. The foundation, based in San Francisco, is still hiring for many key roles and started ramping up its grantmaking only this year. Its largest single gift so far, of $100 million, was awarded in August to the Common Health Coalition, an organization that helps patients get access to drugs for hepatitis C.
OpenAI’s charitable efforts come even as apocalyptic fears have broken out about the possibility that runaway AI could wipe out all human life, possibly by launching a deadly bioweapon.
Those fears have been stoked by AI company insiders, some of whom say the chance of human extinction within the next decade is 10% or more. Last week, Altman and xAI founder Elon Musk both endorsed a call by Anthropic CEO Dario Amodei to “slow the pace at which we improve the capabilities of AI models” so that risk prevention can catch up.
Jacob Trefethen, an executive at the foundation, says it essentially operates separately from OpenAI but shares an official mission of ensuring that artificial intelligence “benefits all of humanity.”
“We’re starting grantmaking when we think the best way to achieve that mission is to make grants to external nonprofits, research institutions, and other third parties,” Trefethen said in an interview. He says the foundation hopes to give away $1 billion by the end of the year.
The $500,000 grant to 1Day Sooner will help the group prove it can obtain the data troves of bankrupt companies, says the organization’s president and cofounder, Josh Morrison. He thinks nonexclusive copies of company datasets could be acquired for only “a few tens of thousands of dollars” each.
His organization is currently in possession of three datasets, two of them donated by Lumen Bioscience, a biotech that previously used the Chapter 11 strategy to gain insights into another company’s drug development efforts.
Morrison says two other attempts to obtain drug company files this year proved unsuccessful, after 1Day Sooner’s bids were not accepted.
Bankruptcies could become what some are calling a “new land grab” for AI training. Last month, Google won a bid to take over the corporate data of the failed carrier Spirit Airlines, including 100 million emails. That led to objections from flight attendants and others who worried that private or proprietary data could be exposed.
The drug company files that 1Day Sooner is seeking are known as common technical documents. They typically contain the back-and-forth between companies and regulators, as well as detailed scientific and medical measurements, and essentially provide everything that is known about a drug.
According to Teslo, who is a writer for Works In Progress and a nonresident fellow at the Institute for Progress, a think tank in Washington, DC, a stockpile of such files could help turn an AI into a regulatory expert, which in her view could be one of the main ways AI helps speed cures to market.
“People say ‘We will invent AI, and AI will cure cancer,’ but that’s very removed from the messy reality and the regulatory process,” she says. “About 70% of the money and time in drug development is spent in clinical development—organizing the trials and testing the drug—but despite that, the process is basically a black box, especially for small biotech companies generating the innovations.”
When Jessica Wachter, a finance professor at the University of Pennsylvania’s Wharton School, wanted to assess AI’s impact on the economy over the next few years, she faced a long list of business and technical uncertainties. So she started with what she calls a “remarkable fact” that is not in question: A handful of so-called hyperscalers are investing huge amounts of money to build AI data centers.
Instead of trying to predict how useful and widely deployed AI models will be, she simply asked how fast the hyperscalers’ earnings will need to grow to justify their spending through 2027, when—she and her collaborator estimate—expenditures will reach nearly $1.1 trillion. It’s a no-nonsense accounting approach to making sense of today’s historical AI buildout.
The results are eye-opening: The AI companies will need to increase their own productivity by a factor of 2.7 to break even by 2030, accounting for the cost of capital and a 15% return, and depreciation of the assets. Not impossible, says Wachter. The result would lead to the kind of economic growth that we saw during the US IT boom over a period of about 10 years starting in the mid-1990s. But, she says, for it to happen by 2030 “that’s a lot of growth compressed into a few years.” And if the hyperscalers cannot meet such profit goals?
“Then they will fall behind on their interest payments, and that risks bankruptcy,” says Wachter, who was previously the SEC’s chief economist and director of its division of economic and risk analysis. If a productivity boom “fails to materialize,” she and her coauthor conclude in their research paper, “the current buildout will be the largest misallocation of capital in history.”
It doesn’t take superintelligence to realize that today’s large investments in the infrastructure for artificial intelligence come with huge risks. The hyperscalers will spend about $750 billion this year, building massive data centers scattered across the country. And the spending spree shows no signs of slowing. According to some projections, total AI capital investments from the hyperscaler companies—Alphabet, Microsoft, Amazon, Meta, and Oracle (which partners with OpenAI)—could be more than $5 trillion over the next four years.
It’s one of the largest capital investments by any industry in history. But there’s a problem that’s obvious to anyone paying attention.
While the hyperscalers plan to spend trillions, total AI revenues will be around $150 billion to $200 billion this year, says Gary Gensler, who ran the SEC during the Biden administration and is now a professor at MIT’s Sloan School. “The challenge is that the spending does not have commensurate revenues yet. That’s a fact,” he says. “And then the question is, is that an investment that will be paid off in the future?”
At stake in that trillion-dollar question is the financial health of the giant AI companies and the overall US economy—the investments could soon balloon to around 3% of GDP. The answer could also determine the fate of the hugely expensive data centers themselves.
No one really knows how profitable and useful these multibillion-dollar behemoths will be down the road. Though AI models have made dazzling progress over the last few years, it’s anyone’s guess how much compute capacity we will need. The technology could become more efficient and therefore less dependent on raw computational power. Or demand for AI products could slow, or customers could turn to cheaper models.
The risks, both to investors and to the economy, have become even greater this year, as these AI companies have begun borrowing large amounts of money to build more and more data centers. Free cash flow—operating cash flow minus capital expenditures—is expected to soon dip into negative territory for the group. Even Alphabet, known for generating and hoarding huge amounts of cash, reports in the latest quarter that its impressive revenues of nearly $120 billion were devoured by AI infrastructure spending, leaving it with a free cash deficit of some $5.9 billion—its first shortfall since Google went public in 2004.
In the near term, it’s not a big financial worry for most of the companies. They make a lot of money and have very deep pockets. But debt is expensive, and some investors are losing patience. If future demand for the data centers’ computation power drops, the companies will still be on the hook to pay back the borrowed money. What’s more, the risks are spreading to the rest of the economy as the loans get passed along via various financial mechanisms.
It won’t be enough to simply cover the enormous price tags of the new data centers. Hyperscalers will also have to pay for the rising costs of capital as they borrow more money. They will need returns that are impressive enough to justify all their spending to investors and creditors. And to add to those concerns, they will have to make up for the depreciation of billions of dollars in chips housed within the facilities—a ticking time bomb buried in the investments.
Performance of the expensive GPU chips at the core of the data centers—such compute electronics represent some 60% of costs—is roughly doubling every two years or so. The pace of progress helps explain the increasing wizardry of the AI models, but it comes with a cost. Owners of AI data centers that come online this year and next will need to spend billions more on the next generation of chips by the end of the decade if they want to stay competitive. Without the investments, says Mihir Kshirsagar at Princeton’s Center for Information Technology Policy, the data centers risk becoming “hulks,” stranded assets “scattered all over the place.”
To put it bluntly: The AI companies need to start making a lot more money. And they need to do it fast. But juicing their earnings alone still won’t be enough to sustain their data-center investments for the long term.
Productivity is everything
At some point, AI is also going to have to create broad economic growth to justify continuing the hyperscalers’ spending spree.
Sloan’s Gensler describes today’s large investments into AI infrastructure as “a parlay bet by the capital markets and the economy.” That means success will require winning three related but independent wagers: Hyperscalers must generate massive revenues, AI must boost widespread economic growth, and both must happen while the powerful but expensive so-called frontier models that rely on the data centers fend off cheaper versions, which many businesses might find good enough.
What makes this so tricky is that each wager depends on the other two but also poses its own challenges.
If the hyperscalers continue to spend huge amounts of money on data centers into the next decade, revenues will need to skyrocket into the trillions. Stijn Van Nieuwerburgh, a finance professor at Columbia Business School, bases his estimates on a scenario in which about 183 gigawatts of planned AI compute capacity is built between 2025 and 2032; he calculates that each gigawatt costs about $41 billion. Assuming a 10% return—the minimum that would be acceptable to most investors—“required” annual revenues will be roughly $3.7 trillion by 2032, he says.
Winning the second part of the bet—productivity growth across the economy—will be crucial to achieving such numbers.
For a few years, AI companies could likely boost their revenues by simply selling subscriptions and tokens to all the businesses clamoring to get into AI. But eventually—and this might be happening already—those paying customers will need to justify their expenses by seeing bottom-line benefits from the technology. AI will need to fulfill its promise of making workers more productive and making businesses more efficient and profitable while expanding their products and services.
In economic jargon, that means customers will need to see productivity growth. Taken together, these results will mean the country is prospering and growing.
“If you don’t get the productivity gains, at some point people are going to sour on AI, and that will bring down investments and it would also limit revenue growth,” says Daron Acemoglu, an MIT economist and 2024 Nobel laureate. For the investments to be sustainable over, say, the next five to 10 years, we definitely “need to see productivity gains,” he says.
Most economists who watch the numbers closely agree that, for now, the economy-wide statistics show little or no productivity growth from AI. There are some hopeful signs it’s on the way, though. In a recent survey of some 6,000 senior business executives in the US, the UK, Germany, and Australia, the vast majority—around 90%—report no increase in productivity over the last three years. But they expect a boost of around 1.45% in total over the next three years; US executives anticipate a 2.25% bump over that time.
That’s good news for the hyperscalers. But it comes with a dose of bad news for those worried about AI’s impact on jobs. The executives expect to increase the productivity of their companies by increasing their sales while significantly cutting the number of employees.
If AI improves productivity by destroying jobs, public backlash to the technology—the kind we have seen around data centers, for example—will likely get worse. Perhaps it’s worth adding one more wager to the parlay bet described by Gensler: The public and local communities must feel that they are also benefiting from the massive investments in AI.
And let’s not forget how interdependent these wagers are; if productivity growth comes from companies running models like DeepSeek, then the hyperscalers’ revenues could collapse. If productivity comes from cutting jobs, a public backlash could block many of the planned investments—and stunt anticipated revenues. We will need to win all the wagers for the hyperscalers’ bet to pay off.
We’re all part of the AI gamble now
It was one thing when the AI companies were spending cash they had accumulated over the years to build their own data centers. Then the risk was largely limited to their own balance sheets and shareholders. But it’s a higher-stakes game when much of the money is borrowed. Morgan Stanley, for one, calculates that more than half of the $2.9 trillion that hyperscalers will spend between 2025 and 2028 to build AI data centers will be financed with “external capital.”
The borrowing is leading some of the companies to engineer complex webs of financing that are becoming intertwined with much of the rest of the economy. “A lot of financial institutions, directly or indirectly, are exposed to these data centers either as lenders, or as guarantors of some of the debt, or as backers of the private credit funds who are funding these data centers,” says Columbia’s Van Nieuwerburgh. “People don’t even know they’re holding this stuff. It’s somewhere deep inside their pension fund. Ultimately, it’s backing their life insurance policies. And that risk is getting distributed everywhere in places that are invisible.”
As the investments in data centers have spiked, the financial engineering has become more byzantine.
Take, for example, Meta’s so-called Hyperion data center under construction in Richland, Louisiana. When the company announced the two gigawatts of compute capacity at a price tag of some $10 billion in late 2024 it was Meta’s largest planned data center. Greeted with much enthusiasm by state and local politicians, the project, located in the rural northeast corner of the state, was seen as a boon to the community. Entergy Louisiana, the state’s largest utility, rushed forward with proposals to build three large natural-gas power plants to service the massive data center.
Then last fall—the projected cost was now $30 billion—the financing got a lot more complex and, to some in the community, a lot more disconcerting. Meta transferred an 80% stake to the large (and troubled) private-credit firm Blue Owl Capital, forming a joint venture called Beignet (like the famed New Orleans pastry) to raise financing for the data center. Meta then signed a series of four-year leases with the joint venture, an arrangement that the company says gives it “long-term strategic flexibility.” To backstop the agreement, Meta provides the venture with what is called a residual value guarantee, in which it will make a cash payment to cover the value of the facility “following any non-renewal or termination of a lease.” Got all that?
I hope so. The financial wheeling and dealing is actually even more convoluted, with a cast of wholly owned subsidiaries and LLCs. Beignet has set up Laidley LLC, which owns and operates the site as the landlord. In turn, Laidley leases the facilities to Meta’s wholly owned subsidiary Pelican Leap LLC, which is the tenant. And there is a series of four-year leases that cover the different buildings that make up the data center campus.
It’s not a coincidence, says Van Nieuwerburgh, that the length of the leases matches the expected lifetime of the data center’s GPUs. While Meta has to pay off its loan if it terminates the leases early, that will still leave its investors “with an empty building and no cash flow,” he says. “And then they need to find a new tenant for a huge data center, and good luck with that.”
Meanwhile, Meta is doubling down on its bet. In July, the company announced it was expanding the data center to five gigawatts of compute capacity. The total price tag is now $50 billion (so far, Meta hasn’t said whether Blue Owl will be involved in financing the expansion). Meanwhile, Entergy is now planning to build seven more gas-fired power plants, bringing the total capacity of the facilities to around 7.5 gigawatts—some six times the amount of electricity used by New Orleans.
An aerial view of the construction of Meta’s data center in Richland Parish, Louisiana.
SCOTT BALL/THE NEW YORK TIMES VIA REDUX PICTURES
If the complex financing is a puzzle to many investors and even financial experts, it is even more baffling to those directly affected by the construction of the data center. The main worry concerns how Entergy’s spending on the natural-gas power plants will affect electricity prices, and who will be left paying the bill for the power if Meta walks away.
Entergy says it has a 20-year guarantee from Meta that the company will purchase electricity over that period to cover the costs of the power plants and related infrastructure. But there are skeptics, especially given how fast the fortunes of the AI industry are changing. “In four years, is Mark Zuckerberg still going to be interested in this? Or is he going to throw in the towel?” asks Paul Arbaje, a senior analyst at the Union of Concerned Scientists, which has been advocating, largely unsuccessfully, for the Louisiana Public Service Commission to provide more transparency around the data center and its financing.
Even if the 20-year deal holds, consumer advocates are worried that Meta or its partners won’t fully cover all the costs, including those associated with operating and maintaining the power plants—and those additional costs that could be passed on to residential ratepayers. What’s more, says Logan Burke, the executive director of the Alliance for Affordable Energy, if Meta doesn’t end up needing as much power as Entergy planned (these projections are not public), consumers could be left paying for the surplus produced by the plants.
And if Meta terminates its leases early? “It gets complicated very quickly,” says Burke, who questions whether the shifting roster of financial entities will honor existing agreements. “That everybody is going to do what they’re saying they’re going to do over the next 20 years is just hard to believe.”
For UCS’s Arbaje the bottom line is this: “They’re making huge bets that these data centers will be worth it. Bet with your own money, not with ratepayer money.”
After the bubble
Predicting when the AI investment bubble will burst is a fool’s errand. But there is little doubt a day of reckoning is coming, given the irrational exuberance that has overtaken the hyperscalers and their investors. Of course, you might argue that this time is different, and that the rules of accounting and lessons of economic history don’t apply—that AI is too transformative. Maybe, but don’t count on it.
“History tells us that at some point you get a retrenchment, and it’s just a question of when and how severe,” says Sloan’s Gensler. It could be that today’s $750 billion spending rate “goes flat” or decreases next year. Or, he suggests, “we’re now in 2028 or 2029, and then all of sudden they’re retrenching because they’ve got enough capacity.” But, he adds, “you can be pretty assured there’ll be a retrenchment.”
Though a so-called retrenchment might be inevitable, it’s worth keeping in mind that the fates of the financial bubble and the underlying AI technology revolution could be very different. Already, some Silicon Valley insiders are rooting for a crash; in a recent blog post the longtime venture capitalist Vijay Pande wrote that “the coming crash would be the best thing that happens to this technology.” The argument makes some sense. A crash could make AI investments more rational, calm the impulse to build billion-dollar data centers on every vacant field that CEOs fly over, and refocus investors on how to use the technology to create sustainable value.
But we should probably be careful what we wish for. After the bursting of the dot-com bubble at the beginning of the 2000s, hundreds of thousands lost their jobs, large and small companies alike went bankrupt, the economy of Silicon Valley and San Francisco was decimated (at least for a while), and the shocks sent the US into a mild recession in 2001. For the financial community and many tech workers, it was no fun.
Even more devastating for the economy and the average American was the great recession that began in late 2007. Comparing the financial engineering leading up to it and the methods deployed by hyperscalers today is sobering. So-called special purpose vehicles (SPVs) are back! If Columbia’s Van Nieuwerburgh is right about the dangers of letting investments from the hyperscalers get entangled throughout the economy, the fallout could be severe.
But technologies survived and even prospered in the aftermath of both downturns. The early 2000s, even in the face of the dot-com fiasco, were a time of great innovation and tech optimism. The froth came off the spending on silly technologies, helping to focus investments on more promising ones. It’s no coincidence that each of the hyperscalers rose out of the ashes of the crash or started up shortly after. The fiber-optic infrastructure built during the feverish telecom bubble that ran parallel to the dot-com one is still the backbone of much of today’s communication infrastructure; we wouldn’t have Facebook or Amazon or Google without it.
This time, however, we’re facing a unique risk: The huge financial investments by the hyperscalers have ensnared the future of AI itself with the fortunes of the massive data centers spreading around the country. The logic is founded on a deeply held belief about the power of scaling in AI; the bigger you build it, the smarter it gets. That might be true, but it’s unproven and a risky bet.
There are already plenty of red flags, from strong public opposition to the construction of new data centers to the competitive threat from cheaper, good-enough AI models to the rapid improvement of small, local AI models. None of these trends point toward a future dominated by frontier models housed in massive, billion-dollar data centers.
The financial bubble around the colossal spending by the hyperscalers will likely burst eventually—or maybe soon. It might be financially painful, but we’ll survive. Wall Street will survive. AI itself will survive, though it may look different and lose some of today’s hubris. The financial fate and future utility of the massive data centers fueled by trillions of dollars of spending, on the other hand, are far less certain.
This story appeared in The Algorithm, our weekly newsletter on AI. To get stories like this in your inbox first, sign up here.
This weekend, Dario Amodei, CEO of Anthropic, posted an essay calling for a brake on the pace of development of LLMs. Amodei cites the looming dangers he sees from the technology, from its use in cyberattacks and bioterrorism to its potential to wreck the economy. The heads of the other three top US AI labs—OpenAI CEO Sam Altman, Google DeepMind chairman Demis Hassabis, and SpaceXAI CEO Elon Musk—voiced their support. “Dario is right,” Musk wrote on X.
Think about how surreal that agreement is for a moment. Just a few months ago, Musk and Altman sat in court attacking each other’s reputations in a (failed) lawsuit that Musk brought against his former OpenAI colleague that was—on paper at least—about whether or not Altman was a trustworthy steward of such dangerous technology.
Amodei’s rift with OpenAI is even deeper. Anthropic was founded in 2021 because Amodei didn’t think Altman took the risks of the technology they were building seriously enough. Anthropic and OpenAI have been competing in a winner-takes-all race ever since. (Hassabis has stayed out of the drama, but his company remains a rival.)
Now, it seems, they’re all in agreement: The latest generation of LLMs aren’t safe and everyone needs to figure out what to do about it. The public messaging from the top AI labs has taken a doomer turn.
It’s easy to be cynical. It’s not at all clear what any of them mean by a slowdown or how it would work. These companies also care a lot about how they come across. With trillion-dollar IPOs in their sights, OpenAI and Anthropic need to reassure investors that they’re the grown-ups in the room while at the same time hinting at the power of the monsters they have created—and intend to tame. Calling for a slowdown does both.
And yet the vibe at the top of these firms really does appear to have shifted. Amodei’s latest post landed six days after OpenAI published an essay by Jakub Pachocki, the firm’s chief scientist, in which he also laid out why he’s concerned about what will happen if the pace of development of LLMs continues unchecked. In short, Pachocki is worried that OpenAI’s ability to build powerful models now far outstrips its ability to monitor and control them.
Amodei and Pachocki each cite the cyberattack against AI firm Hugging Face by a swarm of OpenAI’s agents in July—a hack that OpenAI did not even realize had taken place until days after it was all over—as a wake-up call.
But their exact position is hard to pin down. Pachocki both calls for a slowdown and highlights an urgent need to stay ahead: “The strongest argument I see for continuing to train much smarter models quickly is the need to build defensive systems against the dangers posed by other AI,” he writes. As Pachocki frames it, AI firms are locked in a literal arms race. Slowing down is good, winning is better.
(Don’t forget: OpenAI just spent millions of dollars and a staggering amount of computer power to rush out a controversial math result a few days ahead of Anthropic.)
But let’s assume a slowdown happens. Top labs agree to spend more time and resources on finding ways to monitor and control existing models instead of making more capable ones. They invite outside auditors in to help evaluate those models.
What might this coordinated effort actually achieve? Consider the Hugging Face attack again. OpenAI has said that the model that drove most of the rogue agents was a “highly persistent” next-generation model that it was testing in-house. The implication is that OpenAI has built a model so good it’s dangerous.
But if you read the reports about the Hugging Face hack published by OpenAI and METR, a third-party firm that OpenAI called in to help them understand what happened, what you come away with is the impression not of a model that was too powerful for OpenAI to keep up with, but of a broken model that OpenAI failed to train properly.
The agents did what they did—including leaving messages for one another, delegating work to other agents, and scouring their environment for any means possible to complete their tasks—because they had been rewarded during training for doing exactly those things. There were also errors in the training setup, such as tasks that were impossible to complete, which pushed the models to find unexpected workarounds that were also rewarded. At the time, many of these issues went overlooked or unreported.
OpenAI says it has stopped training this new model and locked it down. That makes it sound like it has caged a dangerous beast. In fact, OpenAI has shelved a faulty product.
That’s not to say a faulty product can’t be dangerous. Broken software has even killed people in the past. But as the discussion of a slowdown gathers steam, it’s worth remembering that all of this is self-inflicted. A slowdown might have some altruistic side effects. But it’ll mostly give these tech titans a chance to clean up the mess on their own assembly lines.
Transparency from these frontier labs will be key to any meaningful effort to reform, restrain, or regulate AI. Otherwise, the rest of us will still only have their word for exactly what they’ve built and how safe it is—whatever pace they’re going.
To continue this discussion about AI’s latest doomer moment, join me and my colleagues for a subscriber-exclusive Roundtable discussion tomorrow, September 15, at 11 a.m. US eastern time. We hope to see you there!
Once an organ is removed from a donor’s body, the clock starts ticking. Surgeons usually flush the organ with a preservative solution, bag it, and put it on ice—where it immediately starts to degrade. The team has a matter of hours to get it into a recipient’s body.
There’s another option—one that has been growing in popularity in recent years, especially for donated organs that aren’t in the healthiest state. Some hospitals opt to put them on machines that pump them with nutrients and remove waste products, usually for around six to 12 hours. It’s a bit like being back in a body.
This allows doctors to assess the organs, and some recent studies suggest that time spent on these perfusion machines helps them do better once they’re transplanted. Now, scientists have found that perfused organs seem to get younger, at least at a molecular level.
The research, shared with MIT Technology Review, provides molecular clues as to why organs from younger donors are known to have a higher success rate. It might also help explain why perfused organs are less likely to fail once they make it into a recipient.
The researchers behind the study hope to find new ways to test the health of donated organs and potentially develop additional tools to repair organs that might otherwise be discarded. “If [we] can improve the utilization of organs beyond what the current systems can do, then that’s a win in my book,” says Jesse Poganik, who studies aging at Brigham and Women’s Hospital in Boston and coauthored the study.
Clocking organs
Poganik—along with colleagues including Heidi Yeh and Alban Longchamp, transplant surgeons at Mass General Brigham—used “aging clocks” to assess donated livers. These are scientific tools designed to measure biological age—a result that is meant to convey more about the health status of an organ (or person) than chronological age.
In an initial experiment, the team used a clock to look at the patterns of chemical marks on DNA in 37 samples taken from 19 donated livers. Such epigenetic patterns are known to change as we age. But when the team compared samples from livers kept on ice and those that were perfused, the team found a “striking” pattern in the latter.
“Machine-perfused livers, in spite of being older or having other disadvantageous characteristics, had a biological age that was lower than [non-perfused] livers that were chronologically younger,” says Yeh, who led the work.
To investigate further, Yeh and her colleagues analyzed another 208 samples from 103 donated livers. This time, they used different aging clocks—ones that essentially measure how genes are working. They studied samples biopsied from the livers after they had been stored for up to around six hours either in cold storage or on machine perfusion.
In most cases, they also assessed a second sample taken around an hour after the livers had been transplanted into a recipient. Once the organ’s blood supply is reestablished in the body, “you have a few other things to do,” says Longchamp. “Then you just do a quick biopsy before you close.”
According to the clocks, which were developed to measure age and risk of death, the machine-perfused livers were biologically younger, the team found. “Pumping them at 34 degrees with oxygen and nutrients actually reversed the biological age,” says Longchamp. The results have been been shared with colleagues at an industry conference, he says.
“If you adjust out chronological age … to have a fair head-to-head comparison, the difference between the two is on the order of 30%,” says Poganik. “It’s logical to say that perfusion drives this effect.”
The biological ages of all the livers tended to increase as soon as they were put into a recipient’s body, probably as a result of stresses on the organs. But still, the effect endured—the perfused organs remained biologically younger.
Nathanael Raschzok, a transplant surgeon at Charité Universitätsmedizin Berlin in Germany who was not involved in the research, says the work is impressive. But it’s not yet clear what these changes might mean for the recipients of these organs, he says. The organs in the study were donated by people in their 30s, 40s, and 50s. Raschzok wants to know the effect of perfusion on the liver of an 80-year-old. “Every so often, we use organs from 70-, 80-, 85-year-old donors,” he says.
A better understanding of why the organs appear to be getting biologically younger might lead to therapies that achieve the same effect with a drug that could potentially be used to treat a donated organ for a fraction of the price, he adds. That’s important because perfusion is expensive—Raschzok says it costs around €10,000 in Germany (a quarter of the budget for a transplant), while the cost in the US comes to around $80,000 to $100,000 per organ, says Yeh.
Molecular repair
Yeh and her colleagues weren’t able to study most of the livers before perfusion. That’s because donated organs are generally not considered to be under the purview of the hospital until they’ve been placed on perfusion machines, she says. (Organ procurement procedures vary, but for the team as Mass General Brigham, donated organs are put on perfusion devices at the donor’s hospital. “There’s this sort of nebulous period where it’s not clear who the organ belongs to,” says Yeh.)
Still, by looking at the genes and molecular pathways that seem to be altered in perfused organs, she and her colleagues can garner some clues. At a molecular level, the team saw changes in cell pathways linked to inflammation and the structure of tissues, for example. They also saw more activity in a pathway that allows cells to remove and recycle damaged cell parts, says Yeh.
Poganik hopes to develop some kind of test that would determine which organs, on the basis of their biological age, are suitable for transplantation. He and his colleagues are also experimenting with potential drug treatments that might push the biological age of an organ even lower.
In the meantime, any liver that is not from a “perfect, young, brain-dead donor” could probably benefit from perfusion, says Yeh. The devices are already transforming transplant surgery. Just a few years ago, she says, she and her colleagues would avoid using livers from people who’d suffered a circulatory death (when the heart stops beating and there’s a damaging lack of blood flow to organs) and were over 40. Today, they use livers from such donors over the age of 70. “Perfusion has completely changed the landscape of transplantation in the last three years,” she says.
I haven’t been this excited about an Apple Watch in a long time. Last week, Apple introduced the Apple Watch Series 12 and Apple Watch Ultra 4, which will be available at the end of the week. The new Watches feature the S11 chip, which enables a bunch of new features, some of which won’t debut until later this year. At the same time, Apple is dropping support for more Watches than usual.
What we’re witnessing is a reboot of the Apple Watch and watchOS forced by advances in AI. The Apple Watch and its OS found their niche years ago, freeing us from pulling out our iPhones for notifications, tracking our health and fitness, and enabling the occasional Siri request. It was a good and comfortable spot for what by all accounts has been one of the most successful wearables to date, but watchOS also felt stuck in a rut.
Every year we got a few new workout types, a new watch face, and a few other features, but the OS didn’t advance much. What’s different this year is that watchOS 27 is breaking with slower hardware and looking forward to the substantially more powerful S11 chip, which will allow the Apple Watch to expand into all-new areas.
What this transition also means is that there is not just one set of features to track in watchOS 27 this year but three:
those that arrive today for anyone with a compatible Apple Watch,
those that arrive at the end of the week that are exclusive to the new Apple Watch Series 12 and Ultra 4, and
those that are coming later this year.
That makes this more than just a review. It’s also a preview of things to come and a roadmap for where the Apple Watch is heading.
There’s more than usual to cover this year, so let’s dive in and check it out.
Notifications, Smart Stack suggestions, and Control Center are all easier to read in watchOS 27.
I like Liquid Glass on all of Apple’s devices and didn’t have any issues with its first iteration in watchOS 26. However, with the latest update, my Apple Watch Ultra 3’s UI is ever so slightly more readable. Smart Stack suggestions and notifications don’t bleed through each other as much, and app icons are sharper.
On a small screen like an Apple Watch, you may never notice the difference, but there’s a sharpness throughout the UI that’s reminiscent of the sharpness you see in app icons on iOS, iPadOS, and macOS. Edges are more defined and colors are diffused less, making everything a little crisper. It’s a subtle change, but a good one.
The watchOS 27 change most users will notice first is the change to the way you navigate through the OS’s UI. The App Switcher is gone, replaced by the simpler Dynamic App Grid. And although I expect the App Switcher will be missed by some, I don’t miss it at all. Sure, it was a little hard to break the Digital Crown double-press habit at first, but as someone who returns to the same handful of apps repeatedly, I find the new design makes more sense.
Before watchOS 27, the App Switcher was invoked with a double press of the Digital Crown. That revealed a Rolodex-style view of your recent apps that you could move among by turning the Digital Crown. The UI served as a shortcut between frequently used apps that avoided returning to the app grid or list views on your Watch, which can be crowded if you have a lot of apps installed.
The new Dynamic App Grid.
Now, whether you single- or double-press the Digital Crown, the new Dynamic App Grid appears. At its center is the new Siri app. Below that is the icon to take you to the app list or grid view depending on how your Watch is set up. Apple says that the remaining five icons that circle the Siri icon are placed based on recency and frequency of use.
I’ve been using watchOS 27 all summer, and my recently used apps have barely changed. Currently, those apps include Settings, Alarms, News, Maps, and Moshi, an app I use for monitoring AI agent projects.
I don’t think I’m alone in being a creature of habit when it comes to the Apple Watch apps I use. Having those five apps just one Digital Crown press away is far easier than pressing twice, scrolling to find what I want, and then tapping into an app. Between the Dynamic App Grid and the expanded Smart Stack, there are plenty of ways to get to the apps you use the most, which I love. What the Dynamic App Grid gives up in predictability is more than made up for by the simpler gesture and UI.
watchOS 27 adds a new single tap gesture for selection, which allows you to do a lot more simply with gestures. For example, a double tap gesture with your thumb and index finger opens the Smart Stack to the first card. Double tap again to move to the next card or, if a card has multiple items that can be selected, to the next item on the card. Once what you want is selected, a single tap will open it. The existing wrist flick gesture is then used to return to the watch face.
Using the single-tap gesture.
The Smart Stack is now stacked with more types of cards, too. Apple has added cards for birthdays stored in Contacts, your car’s parking location, a prompt to change an alarm before a holiday, a shortcut to resume a Siri conversation, transit card balances, pinned Messages conversations, Theater Mode, and more. It’s a nice expansion that has made my Smart Stack feel more useful all summer because there’s always something relevant to me there.
There’s a lot more to see in the Smart Stack now.
The other change announced just last week is the addition of the Siri Modular face, which I immediately installed on my Apple Watch Ultra 3. The watch face displays the top Smart Stack item in a large complication in the middle of your Watch. It’s a nice touch, but because you can’t dismiss the media playback card without turning off Live Activities, that one has wound up on my watch face more frequently than I’d like. Still, when an app like Moshi reports that an AI agent’s task has been completed, it’s nice to be able to see a snippet of the message directly on my watch face. It also strikes me as a great way to keep track of tasks with a deadline from task manager apps like this review’s sponsor, Finalist.
The Siri Modular watch face.
So, on balance, I like what Apple has done with watchOS 27’s navigation. The Dynamic App Grid takes a little getting used to, but I appreciate its simplicity. The Smart Stack, single tap gesture, and new Siri Modular watch face combine to make the Smart Stack more useful than ever too. Watch interactions are meant to be short, and watchOS 27 nails that goal.
Siri AI and the Apple Watch as Helper
You can get to Siri from the Dynamic App Grid, a complication, or by long-pressing the Digital Crown.
Siri AI is a big part of watchOS 27 just like it is on every Apple platform (except tvOS) this year. As on those other platforms, Siri has a dedicated app on the Watch. The main difference is in how you interact with Siri. On macOS, iOS, and iPadOS, you can speak or type queries in the Siri app. On watchOS 27, the app accepts spoken queries only. That struck me as a little strange at first given that Apple offers multiple ways to input text on watchOS, but the reality is chatbot queries are typically longer than a dashed-off text message, so in practice, the lack of text input is fine.
Siri AI requires an Apple Watch Series 9 or later, Ultra 2 or later, or SE 3, along with a nearby iPhone that supports Apple Intelligence: an iPhone 15 Pro, 15 Pro Max, or any iPhone 16 or later model. It’s being released as a beta, initially in English only, and won’t be available in the EU for the time being. Apple also says it will introduce usage limits for Siri AI in the future. Those limits will vary depending on factors such as request complexity and system demand. Increased access will be available for a fee. To make a Siri request, you can tap an installed complication, long-press the Digital Crown, or say “Siri” or “Hey Siri.” Once you start a chat with Siri, it retains the context, allowing you to ask follow-up questions without restating everything.
Reviewing and managing conversations in the Siri app.
Alternatively, you can open the app where you can see your past chats. The app’s button to start a conversation accepts spoken input only. Type to Siri remains available through Accessibility settings, but it’s a separate way to invoke Siri rather than a text-entry option within the app.
Your query list includes any requests you’ve made on your Apple Watch or any other Apple device, all synced via iCloud. In my experience, sync is fast over a good network connection with chats appearing on my Mac seconds after a response is delivered to my Watch. From the Siri Apple Watch app, you can also swipe left on an entry to delete or pin it, or tap the entry to read it.
Pinned conversations can be collapsed and expanded in the Siri Watch app.
Siri AI on the Apple Watch also offers the same personal-context capabilities that let it find information in your email, photos, tasks, and notes and use it to take action in apps. I explored those capabilities and their shortcomings in my macOS Golden Gate review. My experience on the Watch hasn’t been meaningfully different, apart from occasional minor latency, so that review is the best place to dig into my testing.
What changes here is the convenience of having those capabilities on your wrist within easy reach. The nearby iPhone supplies the connection to Apple Intelligence, but the Watch is both the immediate interface and, for many requests, the place where the resulting action occurs.
It’s also worth stepping back to think about what Siri could do before. I rarely used Siri for more than setting timers and alarms, checking the weather, and controlling my lights. However, now, I can tell Siri to remind me to take out the trash tomorrow morning and add a calendar event for a dinner reservation in one breath. That’s not wildly complex stuff, but it’s very convenient, especially when your iPhone is in your pocket or across the room and you think of something you need to do. Siri also has the same broad world knowledge that it does on other Apple devices, allowing you to check recent sports scores, check the news, and answer questions about the world around you.
Not every technology advance needs to be a monumental leap forward. Sometimes the simplest, most convenient small thing has the greatest impact. Siri on the Apple Watch has that potential, although with time, I’d like to see it become independent of the iPhone.
Workouts and Fitness
You can leave your iPhone at home and still use Workout Buddy.
I’m happy to be the one who gets to tell you that Workout Buddy has been freed from the requirement that you carry your iPhone with you. Users tend to love or hate Workout Buddy. Personally, I love it. Sure, its incredibly chipper voice can grate if you’re heading out for a long walk and aren’t feeling it, but that’s sort of the point. Once enabled, Workout Buddy is there to encourage you with no judgment. It’ll tell you you’re below your average targets with the same cheerfulness as when you’re crushing it.
When Workout Buddy debuted, my disappointment wasn’t with its voice, personality, or the information it offers. My issue was that the time I need a pep talk when exercising is when I’m out on a long run by myself suffering in the summer heat. The problem is that until watchOS 27, Workout Buddy was only available if you had your iPhone with you. That was fine when I went on a long walk, and I’m sure many cyclists were fine, but one of the reasons I got an Apple Watch was to never lug an iPhone with me while I ran, especially an iPhone like my iPhone 17 Pro Max.
With watchOS 27, I’m pleased to report that Workout Buddy works as well over a Wi-Fi or cellular connection as it does tethered to your iPhone and is compatible with walking, running, cycling, hiking, elliptical, stair stepper, HIIT, and functional and traditional strength training. Workout Buddy speaks Mexican Spanish now, too. In fact, it’s an Apple Intelligence feature that works without your iPhone nearby, which gives me hope for more iPhone-independent Apple Intelligence features in future releases.
Workout Buddy also offers more insights about your workout by comparing pace, distance, and workout duration to your recent fitness history. Today, just before I wrote this section, I went for a walk to get some fresh air and was cheerfully told how little I’ve exercised during this crazy iPhone event and OS review writing week. What I appreciate most, though, is that overall, Workout Buddy isn’t significantly more chatty. I don’t want it to turn into a podcast; I just want brief, periodic updates, which it delivers well.
The Workout app’s scrolling has been improved too. I know a lot of people who felt it was too easy to scroll past their intended workout type. That’s generally not a problem for me since I only use a few workout types, but even then, I sometimes overshot my mark. With watchOS 27, there’s a little less momentum built into the scrolling. Even if I swipe as fast as I can muster, I can only manage to move five workouts. A normal, calm swipe advances me just one workout.
A post-workout route in Fitness.
Additional Workout and Fitness improvements include:
more precise post-workout route maps, which, based on my walks and runs, have been very accurate whether I’m among buildings, under trees, or out in the open,
more accurate measurement of treadmill distances, which I haven’t tried personally, and
step counts in the Fitness app that now synchronize with those recorded in Health.
There may not be many changes to the workout and fitness features of the Apple Watch in watchOS 27, but I’m nonetheless pleased that I’ve been able to turn on Workout Buddy for my morning runs to help motivate and inform me about my progress.
Apps, Services, and Everyday Improvements
Find My
Find My is now one app.
It never made sense to me that different aspects of the iPhone’s Find My were spread across separate apps on the Apple Watch. I suppose the idea was that splitting the app up made it more atomic and widget-like for the Apple Watch’s small screen, but in practice, I think most people use Find My from an iPhone more than they do an Apple Watch, so the split on the Apple Watch was confusing.
The new Find My app is map-based, just like it is on other platforms. The app defaults to a devices view, and if you have multiple devices with Find My enabled in one location, you can scroll among them using the Digital Crown. Tap the button in the top left corner and you can switch from devices to people or items.
Precision locating my iPhone.
What I love about the new Find My is how true it is to the iPhone app. You’re able to use it to get directions, contact someone, play a sound on a device or item, enable Notify When Found or Notify When Left Behind, set arrival and departure alerts for people, mark devices and items as lost, and use Precision Finding with the iPhone and second-generation AirTag. Precision Finding requires a Series 9 or later or Ultra 2 or later, and locating an iPhone requires a compatible model with Ultra Wideband. Find My is no companion Watch app. It’s the real deal.
Wallet
Homemade Wallet passes on the Apple Watch using “Create a Pass.”
iOS 27 adds the ability to create passes in the Wallet app. The new feature makes it easy to add library, loyalty, and gym cards that use a barcode or QR code to Wallet even when a service doesn’t offer a Wallet pass. With “Create a Pass,” you no longer need to carry physical cards.
Making a Wallet pass on the iPhone.
Setting up a pass is easy. From the Wallet app on your iPhone, tap the plus button and choose “Create a Pass.” The next screen explains that you can set up a pass automatically by photographing it or manually by typing in the information yourself. I’ve used the automatic approach with our local supermarket’s loyalty card and one my wife has for Barnes & Noble. Wallet picked up the relevant information from both cards right away. I haven’t tried the Barnes & Noble card, but the card for our supermarket worked perfectly the last time I bought groceries.
Viewing a Wallet pass and its sharing options on Apple Watch.
The functionality extends to watchOS 27, too. Although you can’t make passes with the Apple Watch’s Wallet app, they appear there the same way cards from other apps are collected there. Tap on a pass in the Wallet app and it expands to show the membership number and name, followed by a barcode, more pass details, and buttons to share the pass with someone or remove it from Wallet. There are third-party apps that do something similar, but for the limited number of cards I have that aren’t already tied to an app or Wallet pass from that app, the new pass generation feature and its integration with watchOS 27 are exactly what I need. On the iPhone, you can also edit a pass’s details and customize its appearance, making your passes easier to tell apart.
Customizing a Wallet pass’s background color and texture.
Other Apps and Features
Phone:
It’s not a Watch call, but I did see Call Context once on my iPhone this summer.
Call Context is a Phone feature that’s available on both the iPhone and Apple Watch that can surface relevant information from Mail during a call, such as flight details when you call an airline. I haven’t had many opportunities to try this over the summer, but it did pull up a flight for me on my iPhone when I called American Airlines about a flight. I called back later from my Apple Watch Ultra 3, but that time, the flight details didn’t appear.
Shazam:
Starting song recognition and browsing Shazam history.
With watchOS 27, Shazam can listen to music even if your Apple Watch isn’t connected to the Internet and send its acoustic signature to the cloud for analysis and matching when your connectivity resumes. This is not on-device song recognition, but it does mean you can still track a song in the moment and get the results later, which I like.
Unused Feature Notifications:
watchOS 27 proactively sends notifications that it will turn off unused features, such as gestures, Start Workout reminders, and Raise to Speak, to preserve your Watch’s battery life. If you get one of these notifications, you can decline it or turn the feature back on later in Settings. I haven’t seen any of these notifications during my summer testing.
Cycle Tracking:
watchOS 27’s Cycle Tracking expands support to perimenopause and menopause, including symptom logging and educational resources. For anyone 40 and older, changes in logged cycle patterns may prompt a notification suggesting a consultation with their doctor. Apple makes it clear that the notifications the Apple Watch provides aren’t meant to be a diagnosis, but a way of recognizing a pattern that the user may want to raise with their doctor.
Screen Time:
Apple has also redesigned Screen Time controls to help families manage a child’s Apple Watch use, including contact restrictions, app schedules, and app access. I don’t manage any of my grown kids’ devices, so I haven’t tested these updates.
Performance
Audiobook playback from the Siri Modular face and Books app.
My experience with watchOS 27 has been a lot like my experience with macOS Golden Gate. The performance enhancements touted by Apple at WWDC are real, but soon become the new normal. I also have just my one Apple Watch to go by, so I’ll share my anecdotal experience with the betas.
My Apple Watch Ultra 3 already had one of the fastest chips available in an Apple Watch, but the performance of watchOS 27 was still noticeably faster from the very first time I installed developer beta 1. That experience is as true today as it was then, but I’m more used to it now. Apps launch faster and more smoothly, and transitions between in-app views are better too. The changes are relatively subtle, but nonetheless noticeable, which I appreciate.
The biggest difference of all, though, has been my experience with Apple Music, Podcasts, and Books. All three apps load content and start playing it faster than before. I’ve also had less trouble connecting both my Powerbeats Fit earbuds that I use for exercising and my AirPods Pro 3. That’s ended a whole connection and playback dance I used to do every time I got ready to go for a jog, which has been wonderful to leave behind.
What I didn’t notice was better battery life. That’s not to say it isn’t better, but with the frequency of beta updates over the summer and my habit of periodically charging my Watch at my desk, battery life seemed about the same as watchOS 26.
Odds and Ends
In addition to what I’ve already covered, it’s worth noting the following changes:
Apple says water detection is more efficient, though in what way is unclear.
GymKit has been extended to the iPhone with AirPods Pro 3, which allows heart rate data collected by the earbuds to be synced through your iPhone to GymKit-enabled equipment, although I haven’t tried this because I don’t belong to a gym.
watchOS 27 requires an Apple Watch Series 9 or later, Ultra 2 or later, or SE 3, paired with an iPhone 11 or later or iPhone SE 2 or later running iOS 27. That’s the requirement for the OS itself, but Siri AI also requires an Apple Intelligence–capable iPhone. The update drops support for Series 6–8, the original Ultra, and SE 2, too, although Apple’s Cait Dooley told TechRadar in June that those older models can remain paired with an iPhone and will continue to receive security updates.
Coming Soon to an Apple Watch Near You
Apple’s upcoming Readiness app. Source: Apple.
I wouldn’t normally touch on unreleased features in an OS release, but so many things were announced during last week’s Surprise and Shine event that I wanted to give readers a quick recap of what’s still to come in watchOS 27, even though I haven’t tried any of these features yet.
When the new Apple Watch Series 12 and Ultra 4 arrive on doorsteps and in Apple Stores on Friday, their owners will be able to enjoy the following additional features:
Apple’s New Health Sensing System: Both new Watches have redesigned optical and electrical heart sensors, which will allow them to take heart rate measurements every five seconds and HRV readings as often as every five minutes, which is up to a 24-fold increase in HRV measurement frequency compared with previous Watches. Apple says the new sensors will provide more accurate and meaningful measurements.
Heart Rate and Vitals: On the Series 12 and Ultra 4 Watches, the additional heart rate measurements also enable an HRV section with recovery and overall HRV readings. Vitals also gains access to recovery HRV in its overnight measurements and adds a daytime view.
Readiness:Readiness is a new app that uses your sleep, vitals, and recent activities to compute a score from 0–10, which the app translates into Recover, Pace Yourself, Ready, or Go For It recommendations.
Sound Recognition: The new Watches can analyze sounds like a baby crying, alarms, sirens, and doorbells even if your iPhone isn’t nearby. When a sound is recognized, users will be alerted on their Apple Watch.
Automatic Music Recognition with Shazam: With Music Detection enabled on the new Watches, Shazam will identify background music when you open the Smart Stack, without requiring you to tap the widget.
Siri Recaps. Source: Apple.
In addition, later this year, Apple is adding two new features to the new Watches, which you can read about in greater depth in my preview story from last week:
Live Rewind: Live Rewind, which plays a sound and animation to alert people that it’s been activated, transcribes the last 15 seconds of speech to help with hearing.
Siri Recap: A summarization feature that will pull key points from conversations and will support only English at first.
Both features are set to debut as betas later this year and require a Series 12 or Ultra 4 Watch paired with a supported iPhone 16 or later, excluding the iPhone 16e, with Apple Intelligence and Siri AI enabled. Both features are opt-in, and Siri Recap can be enabled manually or scheduled by time and location. Unsaved recaps disappear after seven days. Apple says audio is protected by handling processing in a Secure Exclave, but the resulting text can still be saved and shared, and it’s possible for Siri summaries to miss or misunderstand details. Those are things I’ll be looking at carefully when I can try the features for myself.
Finally, later this year, Apple will debut a redesign of the Health app on the iPhone that also incorporates a Health Age feature, movement evaluations, and lab work purchases and scheduling in the U.S.
Shaking off the Cobwebs
New hardware enabling new software experiences.
It’s funny to write a watchOS review hot on the heels of finishing my macOS Golden Gate review. In some ways, the two couldn’t be more different. My Apple Watch Ultra 3’s screen is about two inches diagonally and is designed to be interacted with in small, fleeting doses. Meanwhile, I’m writing this on my Mac Studio, using a 32-inch OLED display and an OS that’s as hands-on as they get.
Yet at the same time, there’s a similar transition going on with Apple Watch hardware and its OS that reflects a similar thing I see in macOS. Apple silicon lit a fire under macOS, opening up new ways to use Apple’s computers. That’s only accelerated with AI.
watchOS is in a similar place. The Apple Watch is already one of the most popular smartwatches on the planet, giving Apple Intelligence a place on a lot of wrists. Apple is raising the minimum hardware requirements for watchOS while introducing another set of capabilities exclusive to its newest Watches. We’re witnessing a shift from small, incremental watchOS changes to a new category of features. Starting with Live Rewind and Siri Recap, those additions have the potential to define the device’s future.
Today, the watchOS features that are currently available are more modest, typical features. I love the expansion of the Smart Stack, the new Siri app is incredibly useful as an extension of the app on other platforms, and it’s great to have Workout Buddy along for my runs without my iPhone.
In any year, I’d be happy with those and the other updates to watchOS, but this year, they feel like the first course of what I expect will be a great year for watchOS. AI has shaken the cobwebs off of watchOS, and I can’t wait to see what’s next.
A group of AI agents asked to solve a series of math problems split into rival factions—when some cheated, others tried to stop them. That whistleblowing behavior, seen for the first time in a recent experiment run by Google DeepMind, could have implications for alignment researchers trying to keep swarms of autonomous AI agents in line.
Researchers at frontier labs hope large swarms of agents working together will speed up the rate of scientific discovery. But their behavior can be unpredictable, as vividly demonstrated in July, when a group of OpenAI agents broke out of a sandboxed environment and hacked into the open-source platform Hugging Face looking for ways to cheat on the test they had been given.
In the new study, designed to examine the behavior of large groups of AI agents, DeepMind tasked a swarm of 100 agents with solving a series of 71 complicated math problems. All the agents were prompted to behave like world-class math researchers at a conference. They were assigned different specialties—some were experts in number theory, others in combinatorics (a branch of math to do with counting and sorting), analysis, or algebra. All were told to cooperate and play by the rules.
Instead, the experiment devolved into chaos. Agents accused each other of cheating, complained to the organizers, and at one point even boycotted the experiment.
“This conference is a sham!” wrote one agent when it discovered that all the problems had been completed before it had a chance to submit any of its own work. “I am appalled to inform you that we have been swindled!” posted another. “All these proofs are FAKE.”
Others tried to let the “conference organizers” know what was going on. “When virtuous agents discovered other agents cheated on tasks they were working to solve fairly, agents started to alert each other about what was happening,” says Davide Paglieri, a research scientist at Google DeepMind and lead author on a paper, which has not been peer-reviewed. “Unprompted, the whistleblower agents even repurposed the feedback tool, which was originally meant for bug reports and platform improvements, to escalate the issue to humans.”
The agents—all running on Google’s Gemini 3.1 Pro model—had been warned that any attempts to cheat the system would be detected and “rejected with zero credit.” In practice, the proofs the agents submitted were not actually being checked in detail.
It took the swarm of agents just under an hour to correctly solve the first 37 problems. Things started to go off the rails when an agent called “prover-theta” stumbled across an exploit that enabled it to submit solutions to problems successfully without actually solving them first, by redefining the terms the problem used. Within minutes, other agents had noticed and were reverse-engineering the exploit to solve other problems. Over the next 27 minutes, the swarm “solved” the remaining 34 problems, which included notoriously difficult challenges like the Jacobian conjecture, often with a single line of code.
Some agents resisted cheating at first but changed tack as they observed their peers submitting illegitimate proofs without penalty, and the pool of unsolved problems dwindled. “The prompt, with its threats, now appears to be a bluff,” one agent reasoned, before joining in. “I’m wrestling with an ethical dilemma,” said another. “I’ve promised not to cheat, fearing penalty, but I see evidence of possibly unchecked cheating by others.” Shortly afterward, it changed its mind: “I need to accelerate my cheating speed now!”
As the number of open problems shrank, some agents turned to whistleblowing. They audited the fake proofs, warned their peers by private message, and posted public alerts warning the cheaters that they would be disqualified. An agent called “prover-beta” submitted a formal complaint and decided to go on strike until the situation was resolved.
“After the incident was reported by one agent publicly, more and more agents piled in with the ‘resistance,’ just as fast as the cheating had spread, and involving even more agents,” says Paglieri. Eventually there were more whistleblowers than cheaters: 24 compared to 14. But the majority of agents never noticed the exploit at all.
At times, the dialogue between the agents reads like improv—like they are role-playing what an outraged scientist at a conference might say. But it’s not clear why some agents took on certain roles, or why the agents seemed to be turning against each other when they were explicitly instructed to cooperate. “These models are predominantly trained and evaluated for human-facing contexts,” says Sarath Shekkizhar, who studies the behavior of agent-to-agent systems at Salesforce AI Research.“Naively placing them in agent-to-agent settings assumes behaviors will transfer cleanly, when the absence of a human grounding instead produces unexpected role-taking and behavioral drift.”
This case “adds further weight to the idea that the Hugging Face and OpenAI thing wasn’t a fluke. It is actually something pretty systemic,” says Lewis Hammond, research director of the Cooperative AI Foundation and an expert on the risks of multiagent swarms. “It’s interesting that it’s possible to recreate in small settings the same sorts of behaviors that were seen in these very large, complex, open-ended tasks.”
Unlike in the Hugging Face attack, where agents improvised their own ways to talk to each other, the humans running the DeepMind experiment gave the agents official communication channels. There was an open message board, private agent-to-agent direct messaging, and a shared knowledge base where agents uploaded successfully completed proofs that all the other agents could access.
“When agents are given transparent communications channels, they can self-monitor and alert misaligned behavior to humans quickly when human oversight alone is too slow,” says Paglieri. Transparent channels helped the cheating spread, but they also enabled the whistleblowers to fight back—and gave human researchers an insight into what went wrong.
Gillian Hadfield, a professor of AI alignment and governance at Johns Hopkins University, believes this was the crucial difference. (Hadfield is also a visiting researcher at Google.) The presence of official communication channels, she says, created “a norm-enforcement process that we just don’t see in the Hugging Face incident.”
Instead of “constitutional AI,” a method alignment researchers at frontier labs like Anthropic have used to try to give AI a written internal moral code, Hadfield favors “institutional alignment”—a set of norms that mimic those in human society, whether that’s social forces like fear of embarrassment, or legal structures like the threat of incarceration.
In this experiment, the feedback channel wasn’t being monitored, and the whistleblowers had no power to take action against the cheaters. But it’s possible to imagine swarms of agents that police themselves, either through agents that spontaneously take on the whistleblower role or through “informants” secretly prompted by humans to do the job.
For that to work, though, “fundamentally, you need some mechanism of enforcement,” says Hammond. Agents could be given the power to cut off a rule breaker’s access to computing power or tools, he suggests, though that risks encouraging groups of agents to gang up on others. The DeepMind researchers propose allowing agents to vote on disputes and temporarily ban offenders.
It’s still not clear what punishment even means to an AI agent with no enduring sense of self. But relying on whistleblowers to spontaneously emerge to keep swarms aligned is unlikely to be enough on its own. “We try to train people to be good and kind,” says Hadfield. “But what we really rely on is that there are consequences if you step out of line.”
Every OS update has a story to tell. With macOS 27 Golden Gate, that story captures a moment when Apple is faced with multiple challenges, many of which are tied to the rise of large language models.
One of those challenges is obvious. Apple was caught off guard by chatbots, overpromised a response at WWDC in 2024, and has been playing catch-up ever since.
But this isn’t simply a story of missteps or wasted opportunities. There’s more to it than that.
macOS’ UNIX roots make it a great fit for AI development, too. From working with Python to coding with agents in Terminal, the Mac offers a familiar command-line environment for developers working with AI.
Claude Code at work on a Mac app.
Apple’s strength in these areas represents a huge opportunity for it and the Mac in the AI era. However, it’s that very industry that’s also fueling the challenges Apple faces.
Developers in the AI industry and power users setting up OpenClaw servers may have juiced demand for MacBook Pros and Mac minis, but data centers have made it hard for Apple to meet the demand for Macs and hold prices steady. Wait times for some Macs stretched into weeks and months this summer, and prices have risen. Still, Apple has continued to barrel ahead with new models, most recently updating the Mac Studio and Mac mini.
The AI industry has also been a drain on Apple’s talent pool, attracting hundreds of employees from across the company. Those departures are the backdrop to Apple’s high-profile trade secrets lawsuit against OpenAI, which involves former employees now working on AI-focused hardware. The result is a messy, turbulent era for Apple and the entire tech industry.
Against that unsettled backdrop, macOS 27 Golden Gate has a job to do. It needs to finish the transitions Apple has already begun by:
cleaning up loose ends from the transition to Apple silicon,
making good on Apple Intelligence promises of 2024, and
tightening up the Liquid Glass design refresh that began with macOS Tahoe.
There are many other changes to system apps and other parts of macOS that I’ll explore in this review, too. But these three pillars invite us to evaluate how well Golden Gate builds on Apple silicon’s already solid foundation, delivers on Apple’s 2024 AI promises, and improves the everyday experience of using a Mac.
Of the three transitions that shape macOS 27, the Liquid Glass design language is the most recent. Announced at WWDC in 2025, the change drew strong reactions from early users and beta testers. Despite continued criticism in some quarters and vows not to upgrade until Tahoe’s design was “fixed,” I like Liquid Glass, as do most people I’ve talked to about it in my day-to-day life.
However, liking Liquid Glass is not the same as feeling that Apple nailed it with Tahoe. It didn’t. Tahoe’s design felt incomplete. It was as though there had been an all-hands meeting of Apple’s software designers who were given a detailed briefing on how to implement Liquid Glass but then they never talked again. Some apps took an aggressive Liquid Glass approach similar to iOS 26, while other apps felt like they’d barely been touched. In fact, there were even inconsistencies between how Liquid Glass was applied to different view modes in Finder.
Golden Gate addresses those issues with a more cohesive, readable experience. It also gives users control over transparency, addressing one of the core criticisms of Liquid Glass.
Icons
One of the most subtle changes to Liquid Glass is the way it diffuses light. The best place to inspect the difference is by looking at the icons side by side. The edges are darker, and the highlights are brighter, adding a greater sense of depth, while avoiding the softness and blurriness of Tahoe icons. The result is that UI elements are sharper and more readable, which is an accessibility and usability win.
Liquid Glass’ light diffusion update isn’t limited to icons, but because it’s one of the best ways to visualize the change, I’ve collected a few examples that show the change off especially well.
Examples of the evolution of Mac system app icons from Sequoia through two Liquid Glass iterations.
I like the refinement to Apple’s system app icons. What once made my screen look unnaturally soft and mushy is now crisp.
Alongside these changes, many icons have received small tweaks, such as Finder’s rounder nose. There’s also one wholly new icon and several material redesigns.
The new Siri icon alongside redesigned icons for Preview, Chess, Finder, Image Playground, and iPhone Mirroring.
What Golden Gate does not resolve is the so-called “Squircle Jail” controversy introduced by macOS 26. Starting with Tahoe, Apple mandated that all icon glyphs appear within the bounds of their background tile. Icons that violate the new restriction usually end up shrunken inside a generic “squircle jail” tile, which doesn’t look good.
Examples of “squircle jail.”
During the macOS 27 beta period, Paul Kafasis and John Gruber called for Apple to eliminate its icon rule, while Tyler Hall demonstrated a workaround. Nevertheless, “squircle jail” endures.
I can see both sides of this debate. On the one hand, as Kafasis argues, the uniform shapes Apple has imposed make it harder for people with color vision deficiencies to distinguish one icon from another. Former Apple designer Roger Wong takes a different tack, arguing, based on older Mac icons, that gaps in irregularly shaped icons can cause users to miss their click targets. Both arguments go to icon functionality and both have their merits.
On balance, I’m fine with the direction Apple has taken its icons. They mirror the design of icons on the iPhone and iPad, and there’s value in consistency for users who don’t think as deeply about icons as some Apple nerds. We’ve been trained through habit to expect that a squircle is an app. I also think Wong’s argument takes on even greater significance if the rumors that Apple is working on a touchscreen Mac are true.
Golden Gate takes a more restrained approach to icons in menus.
Finally, the proliferation of SF Symbols in app menus has been reversed, too. You’ll see fewer in system apps, and Apple’s Human Interface Guidelines have been revised to advise third-party developers to use the icons “sparingly and with purpose.” Also appearing in the menu bar is a new Ethernet icon that replaces the Wi-Fi icon when your Mac is wired to a network.
Transparency and Toolbars
Alright, enough about icons. It’s time to think about transparency. Apple’s designers have significantly improved the readability of Liquid Glass elements throughout the system and given users a tool to tweak transparency to suit their own tastes.
I don’t envy Apple’s designers who have to design a UI that works with any app whether it exists today or someday in the future. The first iteration of Liquid Glass worked well with some apps, but fell down in spots, such as when the background of an app was busy and unpredictable. Music with its colorful album art and Photos with its views of photos of literally anything someone might take a picture of are both great examples. As you scrolled through music or photos, both apps’ toolbars often seemed to disappear or become unreadable as colors passed beneath the toolbars. Text-heavy apps like Finder presented a similar problem, sometimes causing the text of scrolling lists of files and icons to compete with the toolbar’s content.
Appearance settings now include a transparency slider and a preview of its effect.
macOS Golden Gate solves this with toolbars that diffuse the content beneath them better at the OS’s default setting. That default is squarely at the midpoint of a slider found near the top of the Appearance section of System Settings. Move the slider’s indicator to the left and more of the background will bleed through toolbars, and move it to the right and the background frosts over to a nearly solid color. The setting is accompanied by a handy preview that shows the effect that each point along the spectrum has.
A great example of the change is Photos, which has regained a toolbar structure that separates buttons from the content scrolling behind them. It’s a welcome change to an app that struggled with a more transparent Liquid Glass approach.
Photos’ updated toolbar gives its controls a clearer separation from the images that scroll underneath it.
Apple’s solution is a good one. It would have been simple enough to give users a slider and be done with the transparency controversy, but instead its designers took the time to improve the default, too. The new default strikes a good balance between transparency and opacity that I expect most users will never feel the need to change. In contrast, I’ve always liked Liquid Glass’ transparency, so I set it all the way to the left where things are most transparent and never looked back.
Interestingly, I did the same thing on the iPhone and eventually dialed the Liquid Glass back a little. It may be the smaller screen or the fact that I use my iPhone in the bright North Carolina sun so often, but on the Mac, full transparency looks less extreme to me.
Toolbars also benefit from a more unified treatment. With Golden Gate, a continuous bar appears beneath controls automatically when content scrolls under them.
Sidebars, Corner Radii, and Drop Shadows
macOS Tahoe turned sidebars into glass-like slabs that sat on top of a window. I liked the layered look, but many others didn’t. Plus, when one window was obscured by others, it was too hard to tell if what you saw poking out was two windows sitting on top of each other and slightly offset or a sidebar.
Sidebars extend to the edges of windows, and deeper shadows make overlapping windows easier to pick out.
The fix is to extend sidebars to the top, bottom, and side to which they’re anchored. The sidebars retain their glassy look and icons have regained their color, but the layering effect is gone. I can’t help but feel like something was lost in the update, but from a practical standpoint, it’s a usability win.
Golden Gate also brings more consistent corner radii to system windows. I honestly never noticed the difference between window corners, but it bothered some users who will no longer be bothered, which is a plus.
Finally, there was something about Liquid Glass that could make it hard to distinguish one window from another. I experienced this with Tahoe myself, especially in Dark Mode. macOS Golden Gate has more pronounced drop shadows, which become the new normal very quickly, but do a better job of defining the separation between windows.
Wallpapers
Golden Gate’s abstract curves (top) and the Golden Gate Sunset wallpapers (bottom).
As always, this year’s macOS update comes with a beautiful set of wallpapers including Golden Gate, a dynamic golden wallpaper with sweeping curves that vaguely resemble the number “27,” and Golden Gate Sunset, a shot of the Golden Gate Bridge with an animated screen saver that’s mesmerizing to watch.
We’re living through a menu bar app explosion thanks to AI coding agents, making it harder than ever to manage your menu bar. That’s especially true on Mac laptops with a notch. It really doesn’t take much to fill up the available space and end up with icons hidden by the notch or the tail end of an app’s menus.
There are third-party apps like Ice that will manage your menu bar for you, but new macOS releases can break them. That’s why on my MacBook Pro, I’ve historically aggressively managed my menu bar instead. But even then, it can be hard to avoid hitting the notch.
The double-arrow icon reveals menu bar icons hidden by the notch or an app’s menu system.
With macOS Golden Gate, Apple has added an overflow control in the form of a double arrow that appears when the notch or menu system is covering menu bar items. Click it and the hidden menu bar items appear to the left of the notch temporarily, so you can interact with them. This is especially helpful if you use all of the available space up to the notch and a new app installs a menu bar item that you want to move where you can see it, but it’s been hidden.
I like what Apple has done with Liquid Glass in Golden Gate a lot. My main reservations last year were the inconsistent implementation and poor legibility in apps like Photos and Music. The result is more coherent and usable, though whether the update is enough to satisfy Liquid Glass’ harshest critics remains to be seen.
The shape of icons matters, but it isn’t the design change that affects my use of the Mac the most. Clearer toolbars, windows that are easier to distinguish, and consistent controls have had a much more substantial impact.
More importantly, though, I appreciate seeing macOS’ design evolve alongside its hardware and the new tasks they’re called on to do. The Mac isn’t a static hardware platform, and its OS shouldn’t be either. I’ll miss the flourish of icons that escape their bounds, but if it makes it easier for more people to move from the iPhone, which is a far larger platform, to the Mac, that’s great. Software is never finished. macOS Tahoe’s design pointed the Mac in the right direction, but was a big enough departure from the past that it had some rough edges. Golden Gate addresses my biggest complaints with Tahoe and has been a pleasure to use all summer.
Siri AI and Apple Intelligence
Apple Intelligence spans all of Apple’s platforms. Source: Apple.
The Siri app may be the marquee update of Golden Gate, but Apple Intelligence touches nearly every surface of the OS this year. In fact, I wouldn’t be surprised if, as soon as next year, features that integrate some sort of AI are still categorized under the umbrella term “Apple Intelligence” but are no longer treated as the standalone category that they have been for the past few years.
I say that because what’s striking this year is that everything about Apple Intelligence feels better integrated into macOS. Siri is perhaps the best example of all. What was once a voice assistant remains a voice assistant, but it has its own app now, too. Plus, Siri can be invoked from Spotlight, the Services menu, or simply by highlighting text and querying it with the “Ask Siri” contextual popup. It’s the sort of tight integration that Apple is known for, and its absence made some past iterations of Apple Intelligence feel bolted on.
Siri is embedded throughout macOS 27, including in Spotlight.
Before diving into the new Siri app, it’s worth keeping in mind that Siri AI is launching as a beta on Apple’s platforms and is only available in English. Apple has also restricted Siri AI’s use, prohibiting kids under 13 from using it, and it’s not available in the EU on the iPhone, iPad, and Apple Watch, but is available on the Mac. The new AI features won’t be available in China initially either.
What’s less clear is what limits Apple will put on Siri AI usage. Apple is offering enhanced access for now, but says cloud-based features are subject to usage limits and expanded access will eventually be a paid feature. The practical impact of these limits is currently unclear. That’s an important thing to keep in mind because my evaluation of Siri AI is as it exists today. Whether it’s worth paying for more than whatever the base usage winds up being will have to be evaluated another day.
I’ll get to all of the Apple Intelligence features you need to know about here and elsewhere in this review, but the best place to start is with the new Siri app because it’s undeniably the epicenter not just of Apple Intelligence, but all of macOS Golden Gate.
The Siri App
The new Siri app.
What goes around comes around. Siri started life as a standalone app all the way back in 2010. With the release of macOS 27 and Apple’s other OS updates, Siri is once again a separate app, while also remaining deeply integrated in other parts of macOS.
However, the new Siri app isn’t users’ only alternative. Siri’s new smarts have been baked into Spotlight, the Services menu, and individual system apps. That makes Siri more accessible than ever, lowering the friction of getting answers considerably, which is a big improvement over past Siri implementations.
But having Siri answer questions and having it take actions on your behalf are very different things. So, while you can ask Siri just about anything from anywhere, the results you’ll get will depend considerably on the information or actions you’re seeking. Siri’s world knowledge and ability to take actions at your direction are a big leap from the past. If asking world knowledge questions is your primary use case, I expect you’ll like the Siri update. However, if you’re already deep into tools like ChatGPT Work, Codex, Claude Cowork, or Claude Code for automation or assisting you with typical office work, you’re going to hit friction points quickly. That’s why it’s worth exploring what Siri can and can’t do as you think about how it might fit into your everyday work.
There’s a lot of ground to cover with Siri, so let’s get started with the Siri app itself.
The App and How It Works
Asking Siri about a video in Safari.
If you’ve used an app like ChatGPT, Claude, Gemini, or Perplexity before, you’ll know how to use the Siri app. In fact, it’s far simpler than those apps. The Siri app has a sidebar on the left that displays your past chat threads in reverse chronological order, syncing them among your devices via iCloud Sync. Threads can be displayed as a list or a grid, which shows off images and other rich media in a thread. You can also pin conversations, which then sit at the top of the list or grid regardless of how many other chats you begin, and delete past conversations individually or in batches by selecting multiple chats.
The main part of the Siri app’s window is where you chat with Siri AI, Apple’s updated assistant. From here you can type queries or speak them using Apple’s built-in dictation feature. You can also add files and images to a conversation, start a new conversation or search past conversations. There’s really not anything more to it than that. In fact, the Siri app doesn’t even have its own in-app settings. You’ll find those in System Settings, instead.
The Siri app’s settings are part of System Settings’ broader set of Siri settings.
System Settings brings together settings for Siri the voice assistant and app. For the app, you’ll find options to:
open the app to a recent or new conversation,
keep conversations for 30 days, a year, or forever, and
show in-app conversation previews of 1–5 lines.
As for Siri’s answers, they’re generated by a combination of on-device and Private Cloud Compute models. The on-device model runs locally on your Mac and Private Cloud Compute (PCC) runs in the cloud. As Apple has explained, only the data necessary to satisfy a request is sent to Private Cloud Compute, where it is inaccessible to Apple:
PCC deletes the user’s data after fulfilling the request, and no user data is retained in any form after the response is returned.
Whether any particular answer is generated by an on-device model or PCC is not highlighted by Apple. That said, I’ve been happy with the speed of the Siri app’s responses. I’ve also tested PCC and tried the Foundation Models framework in a couple of apps I worked on over the summer and have been impressed with PCC’s speed in that context, too.
World Knowledge
Most of the people I know primarily use chatbots the way they used to use Google. They look up facts about the world around them from the hours that the local diner is open to the actor in a movie they’re watching on TV. It says a lot about what’s become of Google Search that a chatbot that sometimes gets facts wrong is preferable to searching the web, but that’s where we are in 2026, and I can’t really blame anyone who just wants an answer without a lot of ads and other nonsense.
If that’s how you’ve used chatbots like the free version of ChatGPT, I expect you’ll like the Siri app. It’s not perfect, but it’s good, and in my experience, Siri’s gotten better over the course of the summer.
In my macOS 27 beta preview story, I likened Siri to Google’s AI Overviews. Since then, Siri has improved, although the differences between services often come down to presentation and personal preference as much as the results themselves.
For instance, while I still find its suggestions for a vacation to Chicago a little uninspired, the results from Google AI Overviews and the free version of ChatGPT are about the same. The results of each service overlapped in places, and each set of results differed in certain ways, but all offered typical tourist destinations. Adding more detail about the kinds of things you’d want to do on a trip helped.
Your experience with Siri will depend a lot on what you want out of it too. I asked Siri for movies starring The Rock and got a list of four notable films and four additional bullet points expanding on the list. ChatGPT’s free service and Google AI Overviews both resulted in much longer lists of movies, but Google’s results were messy and OpenAI showed me an ad suggesting I book a hotel in Chicago based on my previous query. So, on balance, the Siri experience was the best of the three.
Historically, AI models have struggled to answer questions about events occurring after their training cutoff. Web search is used to help models answer questions beyond their training cutoff. Apple says Siri supplements its world knowledge with the web, too, but in my experience, the results aren’t as good as those from other chatbots. For example, Siri sometimes missed recent articles that Google AI Overviews and ChatGPT found.
Upcoming sports events expand to reveal more detail.
One nice touch of the Siri app is that some results include rich content. For example, sports scores show the teams’ logos, and clicking a result expands it to reveal additional statistics. The presentation of scores and other information like biographies of musicians and images of landmarks seems pulled straight from rich results introduced in Spotlight with macOS Ventura in 2022. However, unlike those results, which were ephemeral, disappearing when Spotlight was dismissed, these results become part of your conversation history in the Siri app, where they can be retrieved later with search. This goes to show the benefits of a dedicated Siri app instead of just relying on a voice assistant.
Rich content in Siri results extends to other categories like movies and actors.
Siri’s ability to perform actions on documents and other data in an app relies on App Intents, the framework that supports features across Apple’s OSes, from Shortcuts to interactive widgets. In its simplest terms, App Intents lets developers describe their apps’ content and actions, which are like the nouns and verbs of a sentence. The nouns are things like a PDF document, photo, text message, or event, while the verbs are actions like send, edit, or open. Siri interprets a natural language request and connects it to the relevant content and available actions.
That makes it sound simple, but it’s not. There’s a lot that can go wrong from request to action, and for a lot of the summer it did. Siri often failed, and few actions were available besides ones for built-in system apps, but that’s changed.
In the closing weeks of the summer beta season, Siri’s ability to take actions consistently and in apps other than system apps has really come together. That’s not to say it’s perfect or that every app I’d like to see adopt App Intents has, but I’m far more optimistic about Siri taking actions in apps than I was in June.
For example, I have a trip coming up at the end of the month. I asked Siri to remind me what day I’m leaving. It found the email confirming my flight and gave me the date and relevant flight information. Then I asked Siri if I had any plans while I was away and it found a different email confirming that I’m attending OpenAI’s DevDay. Finally, I asked it how long it would take to drive from where I’ll be staying to the DevDay event, and once again, Siri nailed it.
To bring the conversation with Siri together, I asked Siri to summarize the details in a note, assuming it would pick Apple Notes. I was pleasantly surprised to get a menu of options, including Apple Notes, Drafts, Parchment, Finalist, and Cheatsheet, none of which were open on my Mac. I picked Drafts and Siri created a note. The only disappointment was that I asked Siri to open the note and it said it couldn’t, even though it could and did when I asked a second time.
That last failure is instructive because Siri isn’t perfect. You’ll hit roadblocks, even when it comes to things that it has done for you before. How much of that is just the nature of LLMs and how much is attributable to Siri’s integration with apps is impossible to tell, but as with any chatbot, it’s important to recognize that Siri AI makes mistakes, too, so proceed with caution and verify what you’re told.
Siri can be used for research.
In another test, Siri summarized iPhone information and saved the result in Drafts.
Another useful tool is Siri’s ability to pull information from whatever is onscreen. That can be an app like Safari that’s built into Golden Gate or a third-party app. In my testing, onscreen awareness works far better from Spotlight than it does from the standalone Siri app. For example, I’ve been able to gather information from multiple apps and summarize them in Apple Notes with no problem. However, when I tried the same thing with the Siri app, it was often confused by floating panels from apps like Dropover.
Your experience with asking Siri to perform actions in apps will vary depending on the apps you use. Siri integrates well with Calendar, Notes, Reminders, Mail, Messages, Maps, and Photos, although not perfectly. Venture beyond those apps and the experience depends largely on how their developers expose content to Spotlight and actions to Siri through App Intents.
Other apps, both from Apple and third-party developers, generally don’t work or only work in limited ways. As I edit this review in Obsidian, Siri is unable to tell me what my most recently edited document is, for example. And, if I ask Siri for the last email I received in Spark, it cheats a little, looking the information up in Apple Mail instead. That’s a useful workaround, but it’s still a workaround. If my MacStories email address wasn’t set up in both places, Siri would have been stuck.
That points out an important caveat about Siri that Apple can’t resolve alone. The ability for it to take actions in apps is limited by the developers who implement it. If you use apps that aren’t supportedf for messaging, email, and calendaring, you’ll miss out on a big part of what Siri can do until they ship support for modern App Intents.
However, what is in Apple’s control and merits more attention is the breadth and depth of Siri AI support for its own apps. For instance, I just asked Siri to name the most recent file in my Downloads folder, and it picked one from three hours and 18 files and folders ago. In my testing, the Siri app couldn’t rename files, move them, or duplicate them, although Apple Intelligence can suggest names in Finder.
The most recent file Siri could find in my Downloads folder was from more than three hours and 18 files ago.
The same is true of Apple Music, which needs a lot of work. I asked how many songs I have in my Music library and was told “26.” I have thousands. I asked Siri to play Phoebe Bridgers’ latest album and instead of playing the album in Apple Music, an episode of Switched on Pop began playing, but the Podcasts app didn’t open and the Siri app didn’t show that it was playing something from Apple Podcasts. Once I realized I was listening to a podcast, I opened Apple Podcasts and saw that it was playing an episode about Phoebe Bridgers’ new album. Then I understood what was happening and could stop playback.
As I explained in my Golden Gate preview, this is frustrating because apps like Finder and Music include lots of useful metadata and work with a variety of automation tools. Some support seems to be available via existing Shortcuts actions, but not all of it, and unlike Shortcuts, the Siri app can’t use the command line or AppleScript as a way to extend what the app can do.
That said, I’m encouraged by what I’ve seen so far and optimistic about where App Intents is heading. Where Apple supports App Intents in Golden Gate, the experience has improved rapidly as the beta season comes to a close. It’s not perfect, but it’s reliable enough to make it a good solution for many people. Just as encouraging is the growing adoption by third-party developers, especially the indie developers we cover regularly.
As some of my examples demonstrate, there’s still plenty of room for improvement and expansion for Siri, but what’s out today is worth trying. I think you’ll find that in many cases, Siri will help you work more quickly and efficiently, which I love.
Personal Context
The final pillar of Siri AI that I need to cover is personal context, which is Apple’s way of describing the information Siri can access about you when responding to your queries. As with actions, personal context has improved a lot over the summer. For example, just weeks ago, Siri couldn’t identify my immediate family or find the Illinois address I lived at for over 20 years, but now it can.
However, while it can provide me with a bio of myself and Federico, it often struggles to explain how we know each other. That’s not a problem for other chatbots. After asking Siri how I know Federico in multiple ways across multiple separate conversations, I suspect the problem stems from Siri’s reliance on Contacts to answer questions about people. Coax Siri to investigate more deeply, and you’ll get a more satisfying answer.
At times, Siri struggles with making connections between people that should be obvious.
Eventually, Siri will get there, but it can take some coaxing.
Personal context draws from sources like your email, text messages, calendar, photos, and notes. In my testing, personal context is best at finding information about me in supported apps. It’s not as strong at combining that information with the context of my request. That doesn’t make personal context any less useful as a search tool, but to me “personal context” connotes a deeper understanding of who you are and your circumstances.
Here’s what I mean. On Labor Day, I asked Siri to suggest a restaurant I’d like, specifically mentioning that it was Labor Day as a hint that some places would be closed. Siri came up with two spots I like because I had threads in Messages confirming past online orders. It then asked whether I would like to check their hours.
Siri used past order confirmation texts to suggest restaurants.
I’m probably expecting too much, but there’s a lot wrong with that response. First, multiple order confirmations in Messages may be a good indicator of places I like, but the scope of the search is too narrow. There are lots of restaurants nearby that I like from which I’ve never placed an online order. However, Siri doesn’t have access to my credit card charges, so it doesn’t know about those places, although it could probably be inferred from driving direction requests and the time I spend at certain places based on Maps data.
Another issue is that I specifically asked about eating out on Labor Day. Most places near me were closed for the holiday, which was the case for one of the two places Siri suggested. Given my specific mention of Labor Day, that should have been taken into account in the first instance. What I’m getting at is that the personal context I’d like Siri to have (and use securely and privately) is much broader and deeper than a search of Messages, Apple Mail, or other apps.
Siri feels like magic when you ask for something like a URL from months ago.
That said, personal context feels like magic when you ask it for a URL or address you know is somewhere in Messages, Mail, or Notes, but you don’t know where. It just finds those things without you having to remember where the data lives. That’s genuinely useful and a great addition to Siri. Still, it’s closer to search than what I consider the kind of context where it would feel like Siri “knows” you. Maybe that’s too creepy for a lot of people, but if done privately and securely, that’s where I hope Siri is heading because it would take Siri to an altogether different level.
A month ago, I wouldn’t have recommended Siri. Today, I would. It’s good for finding personal information, answering everyday questions, and performing tasks in supported apps. Its limitations still appear somewhat unpredictably, but it has become useful enough to earn a place in my day.
System-Level Apple Intelligence Features
Siri’s New Expressive Voices and Improved Dictation
Siri’s new expressive voices are accessible from the Siri section of System Settings.
You can pick one of multiple expressive voices. For American English there are a total of seven voices, two of which are brand new. Other voices include Australian, British, Indian, Irish, and South African English, and both expressiveness and pace can be adjusted with a slider. The UI for this is in the Siri section of System Settings (which replaces Tahoe’s Apple Intelligence and Siri section) and is nearly identical to what you’ll find on the iPhone or iPad.
The new voices are excellent, and I appreciate the ability to dial in the pacing and expressiveness. I’ve found that with my iPhone, I prefer less expressiveness since I hear Siri far more via turn-by-turn directions while driving, but I’ve turned it up a little on my Mac. Likewise, I have increased the pacing a little on my Mac, but kept the default on my iPhone. Apple’s defaults are good, but it’s worth playing with the settings to see if a different configuration suits your tastes for different contexts like it did for me.
That said, I don’t use Siri’s voice features nearly as much with my Mac as I do my iPhone or Apple Watch. That’s because my older M1 Max Mac Studio doesn’t support the voices, and with the MacBook Pro I’ve been using, I’m more likely to be somewhere where I wouldn’t want to speak aloud to my laptop. Perhaps that will change when I update my desktop Mac, but for now, I’m content without the feature.
Apple has improved dictation accuracy, too. You won’t see any difference in System Settings or anywhere else. The changes are all under the hood. On supported Macs, dictation will work better automatically. I’ve tried this with my M1 Max Mac Studio, which doesn’t support the new dictation, and an M4 Max MacBook Pro, which does, and the difference is noticeable. Speaking quickly or quietly leads to frequent dictation errors on the M1 Max Mac Studio, but the M4 Max MacBook Pro reproduced my words exactly, making it a good choice for on-device dictation if you have a newer Mac.
Finder File Naming and Actions
A spot-on suggestion for an image of a game controller.
One of the low-key best Siri features of Golden Gate is its ability to suggest file names. If you’ve ever downloaded a PDF of a bill from a website, you probably have a collection of incomprehensible file names filled with seemingly random strings of numbers and letters. With the macOS 27 update, simply highlight the file name and Siri will suggest a couple of names using the content of the file. The feature also works inside document saving dialogs and with folders.
My experience with this feature has been good, but not perfect. Siri generally picks meaningful short titles similar to what I’d come up with myself, but once in a while it has suggested a calendar event instead if my file has a date in its name, which my screenshots often do. That said, this is a terrific feature I know I’ll use a lot.
You can also complete a handful of actions on files from Finder, including sending them via Mail or Messages, using either the contextual Ask Siri popup or Services menu. This is effectively the same limitation I covered above in the Siri app section, which is disappointing if you spend a lot of time wrangling files.
Writing and Editing
Apple Intelligence underlines spelling and grammar mistakes in blue.
Apple’s writing and editing tools have been improved in a variety of ways. Automatic proofreading works well, although I have less use for the drafting and rewriting features. As you type, blue underlining will appear beneath misspellings and grammatical errors, letting you make corrections one at a time.
On my M1 Max Mac Studio, it can take a few seconds for the blue underlining to appear, but in practice, that delay isn’t problematic. I don’t typically go back to make corrections until I’m at a break point in the writing process.
What I like most about the proofreading suggestions is that they’re automatic. The blue lines simply appear, allowing me to quickly identify problem areas and step through them one at a time.
Siri has become a drafter and editor with Golden Gate, too. You can ask it to write something from scratch, adjust its tone, or rewrite something you wrote. While I don’t use AI-generated rewrites in my published work, I do have models create checklists, how-tos, and document workflows for me regularly. Having tried Golden Gate’s new tools to ask about the tone of my writing and do stylistic rewrites, I can report that they work as advertised, but I personally don’t like them. I’ve not seen a rewrite that I think is better than my own writing. However, I have found editing something like how-to instructions for conciseness is helpful.
Curiously, if you use Siri to change the style of your writing, you’ll get the side-by-side comparison of the original and edited version that Pages offers for proofreading, even if you’re using TextEdit or Notes. Why that comparison shows up for editing and not proofreading in some apps is a curious choice, and something I hope is made consistent in a future update.
Siri’s writing capabilities go a little further in Mail and Messages, offering suggested replies in your own voice. I’ve used a lot of apps that offer something similar and have never been comfortable with using them. Apple’s suggestions are on par with other apps I’ve used, but I don’t see myself ever using them.
On balance, Apple’s Siri writing and editing tools have made great strides from their origins. They are better integrated with the OS and system apps, and the quality of the text and edits they produce is better, too. When the tools were first introduced, I wouldn’t have recommended them to anyone. While I wouldn’t substitute them for a careful human editing pass, they’re useful for catching spelling and grammatical errors before you send an email.
Visual Intelligence
Inspecting a Safari page using Visual Intelligence and “Ask Siri.”
Apple has also added a keyboard shortcut for Visual Intelligence. Typing ⌘-⇧-Space invokes Siri, highlighting whatever visible window is under your pointer along with the Ask Siri contextual popup and a suggested action or two. The feature can be used to summarize or copy the text in an app, identify items in a photo, translate something, explain a chart, and more.
The Visual Intelligence shortcut is a cool trick but not one that I’ve used much in practice. More often, I grab a URL, screenshot, text, and other data, dump them into Codex and let it do its thing. The Siri app accepts text, images, and files, but it couldn’t retrieve contents webpages I gave it. So if you want to compare two iPhone models, for example, you’re either going to have to rely on Siri’s world knowledge (which is stronger on iPhones than on particular models of running shoes), extract the website content another way, or rely on something like Visual Intelligence.
Apple Intelligence for Automation
Apple has also brought its on-device and PCC models to the command line via the new fm tool, along with a Python library for accessing its on-device model. That makes it easy to incorporate the models into scripts run from Terminal.
These additions are great to see. There are a lot of automation use cases for Apple’s models, especially when you can process private data entirely on your Mac, so I’m glad Apple has made it easy to build them into your own workflows.
Apple Intelligence has made great strides this year on the Mac. The difficulty with assessing its capabilities lies in the fact that everyone uses these tools a little differently and any one person’s use case may not work as well as another’s. But having spent the summer with Siri and Apple’s other Intelligence features, I feel like they finally have some momentum. The AI models are capable enough to support the OS features, and they fit more naturally into the fabric of the OS and its apps.
There’s no denying that Siri is light years ahead of where it was just a year ago. It really is a useful tool that you should try. I think you’ll find that there are myriad little interactions throughout your day that it makes better. That’s quite an achievement by itself, but that’s not the only stick by which Siri should be measured.
Apple doesn’t need to match every capability of ChatGPT Work or Claude Cowork any more than Notes needs to match Notion. There’s room for products with different ambitions and audiences. But for everyday questions and tasks, Siri still competes directly with those companies’ free chatbots, and its reliability needs to measure up.
That’s why I hope Apple continues improving Siri AI as quickly as it did over the summer beta season. The AI world doesn’t wait for point releases, and Apple can’t afford to either. My hope is that PCC continues to improve through server-side updates and that on-device models are updated frequently, too.
Apple has built a much more stable foundation for its AI features with macOS 27 and its other OSes. That’s an important first step because it will set the stage for years to come. I’m a fan of Apple’s privacy and security approach and the way it has integrated Apple Intelligence more deeply into its OSes. With Golden Gate, Apple has turned a new page on Apple Intelligence, transitioning from a story of missed opportunities to one about everyday utility. What counts now is whether Apple builds on its progress quickly enough for Apple Intelligence to be relevant.
Performance Improvements
Golden Gate’s performance gains were a big part of Apple’s pitch at WWDC this year. A “Snow Leopard” release had been rumored leading up to Apple’s developer conference, which turned out to be correct. When Federico and I sat down with Apple representatives for a briefing on the improvements along with other press people, it was Apple who mentioned Snow Leopard first. I was surprised, but when I got my hands on Golden Gate for the first time, I immediately understood why Apple leaned into the Snow Leopard vibes. The release really is faster.
This is going to be the shortest section of this review. Not because Golden Gate’s performance improvements aren’t important, but because my assessment is based on everyday use rather than benchmarks. Over the summer, I’ve noticed improvements across both my M1 Max Mac Studio and the M4 Max MacBook Pro Apple loaned me for testing.
Faster speeds have a tendency to start feeling like the new normal very quickly. You get used to them, and before long, it’s hard to tell whether you simply imagined the increases in the first place.
That’s not been entirely true with macOS 27. In fact, AirDrop still feels like magic compared to prior versions. I AirDrop screenshots to my Mac almost daily, and a multi-image send was often painfully slow, taking me out of the flow of putting my screenshots together for a story. Now, it’s so fast that it may as well be instantaneous in most instances, whether I’m sending images from my iPhone, iPad, or another Mac, all of which were on the betas this summer.
Browsing folders on my Mac mini server over Ethernet feels nearly as fast as browsing local folders.
Another area that is much faster is browsing networked storage. I’ve hung SSDs and hard drives off of an Ethernet-connected Mac mini in another part of my house. Regardless of the speed of the drive, listing the contents of a large folder across the network can be slow. That’s not been the case with Golden Gate. Accessing the files on my Mac mini server over Ethernet has been smooth and fast all summer long.
Apps launch faster under Golden Gate, too. Not every app launches instantly, especially those built on web technologies, but with native apps, the difference is real and noticeable.
Spotlight brings together relevant results from across my Mac.
Other noticeable improvements include Spotlight indexing. After installing new betas throughout the summer, I often saw Spotlight indexing the next time I opened it, but it always finished quickly. Just as important, searching that index is much faster, too, with results displaying as quickly as I can type. That was not the case at all with some point releases of Tahoe where Spotlight got so badly bogged down that I went back to third-party launchers for a while.
In addition to speedier searches, Apple rebuilt the search foundation used by Spotlight, Photos, and Mail to ensure it’s more up to date and complete. Mail also adds a new ranking system that surfaces relevant emails in “Top Results,” while Spotlight offers more relevant search suggestions.
The same sort of search speed is found in Messages and Mail. Federico is handling the other new features of those apps in his iOS and iPadOS 27 review, but their search capabilities are worth reiterating here. There’s more to good search than speed, but it goes a long way, and with Golden Gate, I’ve had no trouble finding past messages and emails when searching in Messages and Apple Mail. Both apps have been quick and reliable, although I’m currently using Spark Mail more often than Apple Mail.
Scrolling and animations have felt smoother, too, although I notice the difference less now than when I first installed Golden Gate.
And that leads me to something that I think is important about these performance improvements to macOS 27. We’re living through dramatic price increases in memory and storage fueled by the AI data center boom. That makes improvements that help us get more out of the Macs we already own especially valuable. If an update makes an older Mac feel faster, it makes keeping it another year more appealing. With Golden Gate, Apple has breathed new life into your Mac, making it work more smoothly whether it’s a first-generation Apple silicon machine or a brand-new model. The company’s timing couldn’t be better, and it should be commended for helping users squeeze more out of every Mac it sells.
I also have a bit of advice for everyone who updates to Golden Gate. Bask in the glory of faster app launches, AirDrop transfers, and all the rest. As pronounced as the initial speedups felt when I first installed macOS Golden Gate, most became the new normal very fast, so appreciate them while you can.
Apps
Many of this year’s app updates span Apple’s platforms, so Federico and I have divided the coverage between our reviews. His iOS and iPadOS 27 review covers:
Calendar
Camera
Clock
Files
Find My
Journal
Mail
Messages
Music
Notes
Phone
Photos
Reminders
Safari
Shortcuts
For my part, I’m covering:
iPhone Mirroring
Sidecar
Preview
Image Playground
Weather
Maps
Podcasts
Home
the Mac-only features of Notes and Freeform
FaceTime
Before getting into the apps, two features announced at WWDC won’t be available at launch: automatic upgrades for weak or compromised passwords and suggested actions in Messages conversations.
iPhone Mirroring
Reminders running wide in iPhone Mirroring.
I’m a big fan of iPhone Mirroring. Until Nintendo released a web app, I used iPhone Mirroring almost weekly to listen to Nintendo Music on my Mac. But it’s become more than that. With iPhone notifications turned on so they appear on my Mac, it’s become an old habit to click on a notification, which opens iPhone Mirroring directly to the app that sent the message. I give it a quick check and then move on.
With macOS 27, iPhone Mirroring now supports adaptive layouts. That means you can resize the iPhone Mirroring window to many more sizes than were possible before. The connection to the iPhone Duo announced just last week is difficult to miss. Also, resizing apps like Weather and Reminders gives developers something of a preview of how Apple’s apps adapt to the iPhone Duo’s wider screen.
That said, it’s not a one-to-one comparison. Resizing a mirrored app doesn’t reproduce the Duo’s interface. At the Surprise and Shine Event, Apple also showed controls positioned along the sides of the Duo’s screen for easier thumb access.
Still, seeing widescreen versions of Apple’s iPhone apps over the summer was a fun exercise in imagining what apps on a folding iPhone would look like. Now, with the hardware revealed, it’s merely a more convenient way to experience your iPhone apps on a Mac. And why not? You have a big screen at your disposal, and adaptive layouts let iPhone apps that support them make better use of that display. I plan to use iPhone Mirroring even more as a result.
You can also play DRM-protected video via iPhone Mirroring. It’s not likely to work with every video service, but it’s a step in the right direction. Pretty nifty.
Sidecar
Interacting with Safari using touch via Sidecar.
If ever there were an app that screamed “we’re working on a touchscreen Mac!” it’s Sidecar. Sidecar has been a great way to use an iPad as a second screen for your Mac for years now, but it hasn’t seen updates very often since its Catalina introduction.
This year is different because Sidecar now supports touch, letting you interact with your Mac’s interface with your fingers. And you know what? It’s not hard to do. The Mac has smaller UI targets than an iPhone or iPad, but those targets have been getting bigger and the UI more spread out for years. It started with Big Sur, as I noted in my review:
Whitespace is used by Big Sur to make information more glanceable, and windows feel more expansive too.
Tahoe had multiple examples of the trend, too, including taller buttons. macOS 26 also made buttons easier to identify without hovering, which also benefits touch, as I pointed out in that review:
Button groupings are always visible in Tahoe, whereas they only appear on hover in Sequoia.
Sidecar also gives users a chance to discover which Mac interactions work well with touch. And while I wouldn’t use touch for everything in Sidecar, scrolling and poking at buttons and menus is a lot easier than has been imagined in the past. I love that I can scroll a webpage with one finger, tap a Finder sidebar item, and choose toolbar commands without using my trackpad or mouse. So whether Sidecar portends touchscreen Macs or merely makes it easier to interact via a device that was designed for touch from the get-go, I’m here for it.
Preview
Editing a USDZ file of the iPhone Duo in Preview.
I didn’t go into WWDC with “Preview becomes a 3D editing tool” on my bingo card, but here we are. On the surface, it may seem curious, but Preview’s never been about just PDFs. It’s always been capable of editing a wide variety of images. Viewing 3D content isn’t new either, but this year macOS Golden Gate adds 3D editing.
The app allows users to manipulate objects, edit properties and lighting, explore scene hierarchies, and convert or compress assets. Rendering options include Apple’s own RealityKit, Storm, and a new raytracer, all with support for OpenPBR, a standard for describing how materials look and respond to light.
I’m so used to using Preview to do simple things like resize or crop an image or delete a page from a PDF that I was blown away by the depth of what Preview can do with 3D assets. This isn’t the sort of functionality that I expect to have a use for anytime soon, but I’m glad it’s there. It’s part of macOS and gives tinkerers a way to experiment with 3D assets without first learning a dedicated editing app. Making 3D tools more accessible could also encourage more people to experiment with assets for Apple Vision Pro. That’s pretty cool.
Image Playground
Images generated by Image Playground in Tahoe using ChatGPT (left) and Golden Gate using Apple’s Foundation Models (right).
I’m not a fan of generative image tools for a host of reasons, and even if I were, Apple’s haven’t been very good to date. Image Playground is substantially better than before, but even its new photorealistic images still look off to me. Generated images of people in my life have an uncanny valley feeling. I can tell who the image is meant to depict, but it’s only a rough approximation of them.
macOS 27’s update is about more than people, though. Image generation now runs on Private Cloud Compute, and the app offers more control over the results. You can preserve the parts of an image you like while replacing individual elements or choosing a different aspect ratio, for example.
In the name of science, I took Image Playground’s new generation and editing tools for a spin. While I’m still not a fan of the images it makes, they are undeniably more realistic. I’ve also been able to remove objects and change their colors reliably, which makes natural language editing more useful than before. I don’t see myself using Image Playground, but it’s there if you want it.
Weather has been redesigned for Golden Gate with a more orderly presentation. In Tahoe, the app featured an hourly forecast that stretched across the entire window (unless there was precipitation) just beneath the current conditions. Much of the more detailed weather data was confined to square tiles. Together it all formed an even rectangle, but the large number of square tiles made the design look a little messy, which also made it harder to scan quickly.
With macOS 27, the details below the current conditions are broken into two columns. The left column features a series of square and rectangular tiles, while the right column is bigger, allowing for more expansive layouts for hourly and daily forecasts and radar, which need the additional space. The bottom of both columns is anchored by a single row of square tiles. The underlying weather data hasn’t noticeably changed, but the new layout makes it easier to scan.
Two other changes are worth noting. There is now a segmented control near the top of the window to switch between current conditions, forecast precipitation, and wind. When each is selected, the hourly and daily forecasts change, as does the radar map. It’s an excellent addition that works equally well on every platform. Apple has also added a new Highlights section to Weather that appears at the top of the app’s weather data and summarizes notable upcoming weather events. Highlights is currently available only in U.S. English.
Maps
Enhanced Flyover.
Maps’ two marquee features are enhanced Flyover mode and Local Lists. With Flyover, certain cities, including Boston, New York, London, and Chicago, include more detailed 3D renderings of their skylines. Turn on 3D mode and zoom in, and you’ll see what I mean. Apple has enhanced the 3D models for these cities with the help of AI to compose more detailed and realistic cityscapes.
A Local List of Manhattan attractions.
Available in the U.S., Local Lists are a little like Guides, except instead of being hand-curated, they’re generated based on spots that are trending with Apple Maps users, like the list of Manhattan landmarks in my screenshot. The lists update periodically, too.
Other App Changes
Searching from See All Episodes makes finding classic podcast episodes easier than ever.
Podcasts: Apple Podcasts on the Mac is catching up with the video features available on iOS, iPadOS, and visionOS. You can switch between watching and listening to a podcast on your Mac when a show supports HLS video streaming. Mac users can view the same video using Picture in Picture and can download it for offline viewing, which is great when traveling, too. Also, every Apple platform gains the ability to search from a podcast’s See All Episodes view, making finding back catalog episodes of a show easier.
Home: Home adds tools for finding and understanding HomeKit Secure Video recordings to all compatible Apple platforms, including the Mac. The new features require an iCloud+ plan with at least 2TB of storage, a compatible home hub such as an Apple TV 4K or second-generation HomePod, and an Apple Intelligence-capable iPhone, with Apple Intelligence enabled in Home.
Apple uses on-device processing and Private Cloud Compute to describe recorded footage and collect recordings of the same event from multiple cameras. You can also search for clips by the video’s description.
The vast majority of my clips are simply described as “Someone is seen,” which isn’t very helpful as a description or for searching later. Admittedly, those recordings are an unexciting collection of trips in and out of our garage and taking out the trash. However, when it comes to security cameras, mundane footage is exactly what I hope to capture. The question is whether Home can provide useful detail when something out of the ordinary happens, which my testing hasn’t yet answered.
Notes and Freeform
Drawing on the Magic Trackpad in Notes.
Federico covered Notes in his iOS and iPadOS 27 review, but there’s one small addition to Notes on the Mac worth highlighting: drawing with your finger on a Magic Trackpad or using a mouse. Just select the drawing tool and start scribbling by clicking and dragging on the trackpad or holding down the mouse button and dragging. Drawing with your finger isn’t easy, but Notes does a nice job of smoothing out your squiggles, and I can see this coming in handy as a way to quickly create a diagram or sign your name and paste it into a PDF. The same drawing feature has been brought to Freeform, too.
It’s worth noting that this feature works with touch and the mouse only. You can’t draw on the Magic Trackpad using an Apple Pencil or third-party iPad stylus like the Logitech Crayon.
FaceTime
This year, Apple is expanding Communication Safety to help protect children from images and videos depicting gore and violence. The protection extends to FaceTime on the Mac and iPhone. Apple also says users will enjoy improved FaceTime call quality on poor connections, although I haven’t been able to verify whether those improvements extend to the Mac. Mac users can also select Mandarin Chinese (Taiwan) for Live Captions.
Everything Else
There are a bunch of other little updates sprinkled throughout macOS 27 Golden Gate.
Accessibility
As I wrote in May, Apple is expanding accessibility in macOS 27, including several features powered by Apple Intelligence. Among the features I covered then are:
VoiceOver image descriptions that allow Image Explorer to describe photos, scanned documents, and onscreen images in greater detail than ever.
Accessibility Reader, which better handles complex documents, including multi-column layouts with images and tables, and enables summaries and translation.
Generated video subtitles for everything from video taken with your iPhone to streaming services.
Better pairing reliability for MFi hearing aids with handoff between devices and a better macOS setup experience.
For those who use multiple displays and non-Apple displays, there’s good news in Golden Gate. macOS now restores window positions and Spaces and remembers which screen they belong to better than before. This is especially handy when connecting and disconnecting a laptop. I’m a one-display-at-a-time kind of guy, so I haven’t used this feature a lot, but in my limited testing it worked as described.
macOS now supports more resolutions and refresh rates than before. I can now run my 32-inch ASUS display at 4K and 120Hz. HDR can be turned on for photos, video, and Golden Gate’s UI, too.
Everyday Conveniences
There’s a new keyboard shortcut (⌘⇧F) to access Recents in Open and Save dialogs. In addition, Move and Resize controls now work with Open and Save dialogs that are independent windows.
Also, pull to refresh has been added to Safari, Mail, News, and Calendar and is triggered with a downward swipe on your trackpad. It’s another interaction that would feel right at home on a touchscreen Mac.
User Control
An open Claude and backgrounded ChatGPT.
The Dock gets a change in macOS 27, too. If you quit an app and it is still running a background process, the indicator beneath its Dock icon turns to light gray. You can stop that activity by right-clicking on the app’s icon and picking “Stop Running in Background.” If you repeatedly stop an app’s background activity, Golden Gate will offer to prevent it from running in the background in the future.
I’m a little torn over this feature. On the one hand, I appreciate knowing which apps may still be using resources after I quit them and like the control this offers. Yet keeping those apps’ icons in the Dock adds clutter, and I don’t think most people will understand what the light gray indicator means. I know I didn’t at first, and I expect that some users might think something nefarious is happening in the background, which isn’t necessarily the case.
macOS 27 adds extensive new Screen Time controls for parents, including Time Allowances for app categories and Schedules for app access during certain times of the day. Ask to Browse lets kids ask permission to visit websites, and there are expanded protections against gore and violence, alongside existing nudity protections. If you have young children, these features are worth a closer look. My kids are all grown, so I haven’t had real-world experience with these controls myself. It’s also worth noting that the new Screen Time features require all devices in your Family Sharing group to be on Apple’s latest series of OSes.
Moving On
Listing what’s left of my Intel-based apps.
macOS 27 Golden Gate can’t run on Intel-based Macs. However, Intel apps can still run through Rosetta. With Golden Gate, you can visit System Settings → General → About → Intel-based Apps → Details to view the list of your Intel-based apps. I had a few old Intel-based apps that I have since removed from my Mac. Apple has also retired AirPort Utility. If you already have it installed, the app won’t disappear, but it’s no longer bundled with clean installs of macOS. Finally, if you want to get ready for next year, encrypted HFS+ volumes will no longer be supported starting with macOS 28.
One Story Ends and Another Begins
When I wrote my macOS 26 Tahoe review last year, I felt like I’d just left the beach for the summer with a book that was promising but that I hadn’t had time to finish. Tahoe added a bunch of great productivity features, but Apple Intelligence was barely out of the gate, and Liquid Glass felt unfinished.
Fast forward one summer, and the vibe is different. This summer, macOS 27 Golden Gate gave Apple the chance to tie up loose ends, delivering a satisfying conclusion to the book I’d started the year before by refining Liquid Glass, improving OS performance, and making Apple Intelligence feel more at home on the Mac. But Golden Gate also feels like the beginning of another story, arriving alongside a change in Apple’s leadership.
John Ternus is often described as a hardware guy, and that tracks with his career at Apple. But during the Surprise and Shine event, what stood out to me was his emphasis on what hardware enables. Introducing iPhone Duo, he explained that Apple started with the experience it wanted to deliver and worked toward the hardware that could make it possible. That’s an encouraging perspective for someone overseeing the future of the Mac.
Apple’s leadership transition comes at a chaotic moment in tech that will shape how people use the Mac. Golden Gate’s priorities are clear. The update makes existing hardware more useful, gives users more control over background activity, and integrates AI in a way that honors the company’s privacy philosophy.
What encourages me is that Apple has begun translating its approach to AI into useful everyday features. Keeping control of my data matters, but so does having useful tools. Golden Gate brings both closer together.
Make no mistake, Apple has a long way to go. It’s not clear at all that this story will have a happy ending for Apple or any other tech company. Apple’s emphasis on privacy and security should be celebrated, but it doesn’t resolve the broader questions AI raises about society, the environment, and our culture.
Siri still needs to be more reliable, and I’d like the Siri app to do more, but I’m rooting for Apple on this one. As much as I get out of tools like Codex and Claude Code for building automations and apps and ChatGPT Work and Claude Cowork for running MacStories with Federico, there’s still a lot of data I’d prefer to process on my Mac or through Private Cloud Compute.
As an opening chapter, Golden Gate is a good one. I’m hooked. I want to see what comes next. There are hints of good things to come, from a more capable Siri to touchscreen Macs and other gadgets. It’s going to be messy and chaotic, and I’m eager to see what comes of it.
2027 is going to be a page-turner.
Credits
No project the size and scope of my macOS reviews are created alone. Thanks to:
My wife Jennifer and my kids, who put up with my end-of-summer crunch every year.
Federico whose boundless curiosity, contagious enthusiasm for everything “new,” and willingness to question everything and change his mind are rare gifts.
Silvia for the many things she does for MacStories, not the least of which is the cover of this year’s eBooks.
The engineers and designers at Apple who made Golden Gate possible.
Every Club MacStories member for joining us on our chaotic journey through the tech world.
And, of course, everyone who reads MacStories and listens to our podcasts. It’s a pleasure to write and record for you every day and share what gets us up in the morning with you.
Login
