Thinking Machines Lab, the San Francisco startup founded by former OpenAI CTO Mira Murati, has released Inkling, its first general-purpose AI model. The launch adds another US-developed entrant to an open-weight market where Chinese developers produce several leading coding and reasoning models.
Inkling uses a mixture-of-experts architecture with 975 billion total parameters, of which 41 billion are active during processing. It supports a context window of up to 1 million tokens and was pretrained on 45 trillion tokens spanning text, images, audio, and video. Thinking Machines said it also trained the model for coding, tool use, and multimodal tasks.
The release follows the October 2025 launch of Tinker, Thinking Machines’ first product and an API-based platform for customizing AI models. Developers can fine-tune Inkling through the platform.
In a June 2026 assessment, AI model routing platform OpenRouter highlighted DeepSeek V4 Flash, GLM 5.2, MiniMax M3, and Nvidia Nemotron 3 Ultra as four notable open-weight models. Nemotron was the only US-developed model in the group.
Performance and developer access
Thinking Machines Lab’s benchmark table shows mixed results. Inkling scored 77.6% on SWE-Bench Verified, behind DeepSeek V4 Pro and GLM 5.2 but ahead of Nvidia Nemotron 3 Ultra. It also recorded 74.1% on MCP Atlas, 77.1% on BrowseComp with context management, and 79.8% on IFBench.
Thinking Machines said Inkling’s result used a bash-only harness, while the comparison figures were reported by the competing models’ developers.
The model includes a reasoning-effort setting that developers can adjust from 0.2 to 0.99. Thinking Machines said the setting allows users to balance performance against the number of generated tokens. In the company’s testing, Inkling matched Nemotron 3 Ultra’s Terminal Bench 2.1 score while generating about one-third as many tokens.
Developers can fine-tune Inkling through Tinker using context lengths of 64,000 or 256,000 tokens and test it through the Inkling Playground. The model is available through APIs from Together AI, Fireworks, Modal, Databricks, and Baseten. It is also supported by inference software, including SGLang, vLLM, TokenSpeed, llama.cpp, and Hugging Face Transformers.
Inkling’s full weights are available on Hugging Face as the original checkpoint and as a quantized NVFP4 checkpoint. Thinking Machines also previewed Inkling-Small, which has 276 billion total parameters and 12 billion active parameters. The company said it would release the smaller model’s full weights after completing testing.
Enterprise impact
Inkling’s differentiation lies in its open weights, multimodal capabilities, controllable reasoning, and integration with Tinker, rather than benchmark leadership, according to Biswajeet Mahapatra, principal analyst at Forrester.
“Enterprises are most likely to benefit in workloads where domain adaptation matters more than generic model performance, including knowledge-intensive copilots, multimodal customer service, document understanding, operational workflow automation, and agentic tasks that require organization-specific data, policies, and processes,” Mahapatra said.
Inkling’s US origin could also influence adoption among Western enterprises, according to Pareekh Jain, CEO of Pareekh Consulting. He said many Western organizations face regulatory or procurement barriers when considering Chinese-developed AI models.
“Inkling gives those organizations a US-developed open-weight option that they can deploy on their own infrastructure,” Jain said.
However, the benefits will need to be weighed against the cost of deploying the full model.
Running Inkling on private infrastructure requires a GPU cluster with at least 2 TB of aggregated VRAM for the BF16 checkpoint, according to the model card. Thinking Machines lists configurations of eight Nvidia B300 GPUs or 16 H200 GPUs. A quantized NVFP4 checkpoint lowers the requirement to at least 600 GB and can run on four B300 GPUs or eight H200 GPUs.
“Because Inkling is a massive model with 975 billion total parameters, running the full model still requires significant GPU infrastructure, making closed-model APIs more economical for many organizations,” Jain said.
Jain said Inkling-Small may be a more feasible option for many enterprises because it could reduce infrastructure costs and latency while retaining useful performance across key workloads.
Safety and governance
Thinking Machines said it trained Inkling for calibration, instruction following, and resistance to censorship. The company said the model showed “strong patterns of censorship non-compliance” when evaluated by Cognition on its Propaganda and Censorship Eval.
Inkling scored 98.6% on StrongREJECT, which Thinking Machines described as a test of whether models refuse unambiguous harmful requests.
The model’s safety behavior should be retested after an enterprise customizes it, according to Jain. “Model fine-tuning can weaken safety filters, so companies should retest safety after customizing the model rather than assuming it stays safe,” Jain said.
He added that self-hosted and modified versions could diverge from Thinking Machines’ official model over time without receiving automatic updates.
“CIOs need to ensure every AI agent action is logged, auditable, and governed by human approval for high-risk tasks,” Jain said.
Thinking Machines Lab, the San Francisco startup founded by former OpenAI CTO Mira Murati, has released Inkling, its first general-purpose AI model. The launch adds another US-developed entrant to an open-weight market where Chinese developers produce several leading coding and reasoning models.
Inkling uses a mixture-of-experts architecture with 975 billion total parameters, of which 41 billion are active during processing. It supports a context window of up to 1 million tokens and was pretrained on 45 trillion tokens spanning text, images, audio, and video. Thinking Machines said it also trained the model for coding, tool use, and multimodal tasks.
The release follows the October 2025 launch of Tinker, Thinking Machines’ first product and an API-based platform for customizing AI models. Developers can fine-tune Inkling through the platform.
In a June 2026 assessment, AI model routing platform OpenRouter highlighted DeepSeek V4 Flash, GLM 5.2, MiniMax M3, and Nvidia Nemotron 3 Ultra as four notable open-weight models. Nemotron was the only US-developed model in the group.
Performance and developer access
Thinking Machines Lab’s benchmark table shows mixed results. Inkling scored 77.6% on SWE-Bench Verified, behind DeepSeek V4 Pro and GLM 5.2 but ahead of Nvidia Nemotron 3 Ultra. It also recorded 74.1% on MCP Atlas, 77.1% on BrowseComp with context management, and 79.8% on IFBench.
Thinking Machines said Inkling’s result used a bash-only harness, while the comparison figures were reported by the competing models’ developers.
The model includes a reasoning-effort setting that developers can adjust from 0.2 to 0.99. Thinking Machines said the setting allows users to balance performance against the number of generated tokens. In the company’s testing, Inkling matched Nemotron 3 Ultra’s Terminal Bench 2.1 score while generating about one-third as many tokens.
Developers can fine-tune Inkling through Tinker using context lengths of 64,000 or 256,000 tokens and test it through the Inkling Playground. The model is available through APIs from Together AI, Fireworks, Modal, Databricks, and Baseten. It is also supported by inference software, including SGLang, vLLM, TokenSpeed, llama.cpp, and Hugging Face Transformers.
Inkling’s full weights are available on Hugging Face as the original checkpoint and as a quantized NVFP4 checkpoint. Thinking Machines also previewed Inkling-Small, which has 276 billion total parameters and 12 billion active parameters. The company said it would release the smaller model’s full weights after completing testing.
Enterprise impact
Inkling’s differentiation lies in its open weights, multimodal capabilities, controllable reasoning, and integration with Tinker, rather than benchmark leadership, according to Biswajeet Mahapatra, principal analyst at Forrester.
“Enterprises are most likely to benefit in workloads where domain adaptation matters more than generic model performance, including knowledge-intensive copilots, multimodal customer service, document understanding, operational workflow automation, and agentic tasks that require organization-specific data, policies, and processes,” Mahapatra said.
Inkling’s US origin could also influence adoption among Western enterprises, according to Pareekh Jain, CEO of Pareekh Consulting. He said many Western organizations face regulatory or procurement barriers when considering Chinese-developed AI models.
“Inkling gives those organizations a US-developed open-weight option that they can deploy on their own infrastructure,” Jain said.
However, the benefits will need to be weighed against the cost of deploying the full model.
Running Inkling on private infrastructure requires a GPU cluster with at least 2 TB of aggregated VRAM for the BF16 checkpoint, according to the model card. Thinking Machines lists configurations of eight Nvidia B300 GPUs or 16 H200 GPUs. A quantized NVFP4 checkpoint lowers the requirement to at least 600 GB and can run on four B300 GPUs or eight H200 GPUs.
“Because Inkling is a massive model with 975 billion total parameters, running the full model still requires significant GPU infrastructure, making closed-model APIs more economical for many organizations,” Jain said.
Jain said Inkling-Small may be a more feasible option for many enterprises because it could reduce infrastructure costs and latency while retaining useful performance across key workloads.
Safety and governance
Thinking Machines said it trained Inkling for calibration, instruction following, and resistance to censorship. The company said the model showed “strong patterns of censorship non-compliance” when evaluated by Cognition on its Propaganda and Censorship Eval.
Inkling scored 98.6% on StrongREJECT, which Thinking Machines described as a test of whether models refuse unambiguous harmful requests.
The model’s safety behavior should be retested after an enterprise customizes it, according to Jain. “Model fine-tuning can weaken safety filters, so companies should retest safety after customizing the model rather than assuming it stays safe,” Jain said.
He added that self-hosted and modified versions could diverge from Thinking Machines’ official model over time without receiving automatic updates.
“CIOs need to ensure every AI agent action is logged, auditable, and governed by human approval for high-risk tasks,” Jain said.
It’s not so much generosity that’s behind Anthropic’s decision to extend free access to its most advanced model, Fable, for paid subscribers until July 19, analysts say. Its a last-minute move to grab users, data and model evaluation results.
After the free-access period, Anthropic plans to convert Fable to a pay-per-use model, at $10 per million input tokens and a whopping $50 for 1 million output tokens.
That is double the price of its next most advanced model, Opus 4.8, for input and output tokens. “We’re extending Claude Fable 5 access on all paid plans, as well as keeping Claude Code’s weekly rate limits 50% higher, through July 19,” Anthropic’s team said in a July 12 tweet.
Anthropic keeps extending Fable because it does not yet know what its flagship is worth, said Sanchit Vir Gogia, principal analyst at Greyhound Research. “A vendor confident in its price does not move the same cutoff twice in six days, both times at the wire,” Gogia said.
Anthropic is essentially pushing deadlines to test its products, while users gain by being able to put their toughest tasks to Fable, Gogia said.
Anthropic, which did not immediately reply to a request for comment about the situation, has already seen plenty of action with Fable and its sister model Mythos. Both have been touted as the company’s most advanced models yet.
Fable stumbles, then reappears
Fable was officially launched June 9. Just three days later, on June 12, the US government put export controls on it after Amazon researchers bypassed Fable’s safeguards, prompting the model to identify software vulnerabilities and demonstrate an exploit.
Fable’s freebie extension comes after OpenAI’s latest model, ChatGPT 5.6 Sol, became generally available July 9. Sol is cheaper at $5 per one million tokens input, and $30 for 1 million output tokens.
Anthropic and OpenAI are competing aggressively to build market share, said Jack Gold, principal analyst at J. Gold Associates. “Anthropic and OpenAI are looking to go public and the more users they have, the more attractive it is — even if they are not yet producing income,” he said.
In some ways, the two companies are following a well-trodden path to get customers hooked on their products and turned into paying customers. That’s what Meta, Google and Microsoft, for instance, have done over the years with various “free” offers that later morphed into paid products.
Plus, said Gold, ”The more users you have, the better you can train your models across multiple data sets.”
That’s a potential boon for proprietary large language model (LLM) vendors offering free tokens in a bid to lock enterprises and vendors into their AI environments. But numerous experts have warned enterprises not to fall for that tactic. Instead, they argue enterprises should diversify AI development across multiple AI and cloud vendors, and adopt open-source models.
An LLM space race?
According to LLM benchmarks maintained by Artificial Analysis, Fable is the most intelligent model currently available, with Sol just behind it in second place. One benchmark by LiveBench places Sol as being better in reasoning, with Fable better at math, data analysis, instruction following and language. Both models have advantages in coding.
Meanwhile, Cursor and SpaceXAI on July 8 unveiled Grok 4.5, which the companies said can “handle difficult, long-running tasks that require creatively using tools to solve problems, whether in software engineering, data science, finance, legal work, or anything else you do on a computer,” the company said in a blog entry.
Its pricing is even more aggressive than Fable and ChatGPT 5.6 Sol. Grok 4.5 charges $2 for 1 million input tokens and $6 for 1 million output tokens.
There are growing concerns about tokenmaxxing, where enterprises rack up billions of dollars in token spending, blowing past usage limits before finance controls are implemented.
Enterprises might decide to spend more on models such as Mythos and Fable — if the benefits are tangible, said Max Leaming, head of data science and AI solutions at ManpowerGroup. Fable and Mythos may “actually be less expensive to use in spite of the spiked token cost because it’s far more efficient,” he said.
A company might find that the models use fewer tokens, are faster, and can reduce compute time, he said. “Even though the per-token costs may go up, we may see overall costs go down,” Leaming said.
A year ago, most enterprise AI systems generated recommendations. Today, AI systems are approving transactions, routing shipments, updating records, interacting with customers, and triggering downstream software actions with little or no human involvement.
For CIOs, that shift changes the central governance question. The challenge is no longer simply whether an AI model is accurate. It is whether the organization can explain, audit, and defend the decisions the system makes.
When an AI assistant suggests a meeting time or summarizes a document, mistakes are inconvenient. When an autonomous AI system issues a refund, reprices a product, modifies a customer record, or initiates a financial transaction, mistakes carry operational, legal, and reputational consequences.
When those consequences arrive, “the model decided” is not an acceptable explanation.
This is the accountability gap emerging at the center of enterprise AI adoption. Organizations are deploying increasingly autonomous systems while relying on technology that often provides little visibility into how decisions are made. The result is a growing mismatch between the level of authority organizations grant AI and their ability to understand or justify its actions.
Black-box AI may have been acceptable when AI primarily generated predictions. It becomes far more problematic when AI begins taking actions on behalf of the business.
The lesson software already learned
Fortunately, the technology industry has faced a similar challenge before.
As enterprise software systems became more distributed and complex, troubleshooting failures became increasingly difficult. Engineers could no longer rely on intuition to understand what happened when something broke. The solution was observability: the practice of instrumenting systems so their internal state could be understood through logs, metrics, traces, and monitoring.
The goal was not to predict every possible failure in advance. It was to create enough visibility that teams could reconstruct what happened after the fact and identify the root cause.
Enterprise AI now requires a similar discipline.
But AI observability must go beyond traditional software observability. It is not enough to know what action occurred. Organizations also need visibility into why the system believed that action was appropriate.
An auditable AI system should be able to answer questions such as:
What information did the system rely on?
Which tools or data sources did it access?
What alternatives did it consider?
What verification steps were performed?
How confident was it in its conclusion?
What events led to the final action?
These questions are rapidly becoming essential operational requirements rather than technical nice-to-haves.
Why visibility matters more as AI gains autonomy
As AI systems become more autonomous, failures become harder to detect and diagnose.
A human reviewing a single AI-generated recommendation can often spot obvious mistakes. A network of AI agents coordinating multiple tasks across business processes presents a different challenge. Decisions can build upon one another. A flawed assumption early in a workflow can propagate through subsequent actions, creating confident but incorrect outcomes.
The challenge is rarely identifying that something went wrong. Eventually, an error surfaces through a customer complaint, a failed transaction, an audit finding, or an operational disruption.
The challenge is determining why it happened.
Which information influenced the decision? Which tools were consulted? Which safeguards worked as intended? Which ones failed?
Without visibility into the reasoning process, troubleshooting autonomous AI workflows can become significantly more difficult than debugging traditional software systems.
For CIOs responsible for enterprise reliability, compliance, and governance, that lack of visibility creates unacceptable operational risk.
Moving toward glass-box AI
The answer is not to slow AI adoption. The answer is to make AI systems observable.
Increasingly, organizations are seeking AI systems that behave more like a glass box than a black box. The objective is not to expose every parameter inside a neural network. Rather, it is to provide a clear, auditable record of how decisions were reached and why actions were taken.
The most promising approaches share two common characteristics.
The first is verification. Instead of treating a single model’s output as ground truth, systems incorporate independent validation steps before actions are executed. Multiple agents, external checks, business rules, or verification workflows help identify errors before they become operational incidents.
The second is explainability. Effective systems maintain a decision trail that captures inputs, intermediate reasoning steps, tool usage, verification activities, and outputs in a form that human reviewers can understand.
Together, these capabilities create something that has long been expected of human decision-makers but is often missing from AI systems: the ability to show your work.
The regulatory and business reality
The push toward AI observability is not being driven solely by technologists.
Regulators increasingly expect organizations to demonstrate oversight of automated decision-making systems. Emerging AI governance frameworks place growing emphasis on transparency, traceability, accountability, and human oversight.
Customers are moving in the same direction. Whether the decision involves pricing, service, eligibility, or support, people increasingly want the ability to understand and challenge outcomes that affect them.
The result is a convergence of operational, regulatory, and market pressures around a single requirement: organizations must be able to explain what their AI systems are doing.
Three questions every CIO should ask
Before deploying autonomous AI systems, technology leaders should be able to answer three basic questions:
Can we reconstruct the complete decision path that led to an action?
Can we verify critical outputs before actions are executed?
Can a human auditor understand why the decision occurred?
If the answer to any of those questions is no, the organization may be granting more authority to AI than it can responsibly govern.
Accountability will become a competitive advantage
The organizations that succeed with autonomous AI will not necessarily be those that automate the most processes or deploy the largest models. They will be the organizations that combine automation with accountability.
Black-box systems made sense when AI primarily generated predictions. As AI increasingly acts on behalf of businesses, customers, and employees, visibility becomes essential.
The future of enterprise AI will belong not to systems that merely act, but to systems whose actions can be examined, understood, and trusted.
—
New Tech Forum provides a venue for technology leaders—including vendors and other outside contributors—to explore and discuss emerging enterprise technology in unprecedented depth and breadth. The selection is subjective, based on our pick of the technologies we believe to be important and of greatest interest to InfoWorld readers. InfoWorld does not accept marketing collateral for publication and reserves the right to edit all contributed content. Send all inquiries to doug_dineley@foundryco.com.
On the first day of Software Engineering 101, you learned about the SDLC — the software development life cycle. You learned that there is a whole lot more to producing quality software than writing some code and deploying it to production. We’ve studied the SDLC every which way, and we know where and how time is spent within it.
The coding is the most interesting part of the SDLC. It is one of a number of steps in the process, but it’s the part that takes up the biggest chunk of the SDLC schedule.
Agentic coding has changed all of that. I wrote earlier this year about the impacts upstream of coding when writing code is no longer the bottleneck. Here I want to think about the impacts for developers downstream — what we can expect to happen after the code gets written — when agents write the code.
Downstream from coding
Even before agentic coding, we developers spent far more time reading and maintaining code than we spent writing it. Maintaining the code and supporting the product has always been the major lifetime cost of the project as a whole. Now that the bottleneck has moved away from the coding process, we will look for ways to apply AI agents downstream of writing code. So, just as agentic AI has made writing code trivial, we can expect it to improve the remaining 60% to 70% of our job as well.
Maintenance of code can be a real challenge. We are all familiar with how it often goes. We get a vague customer report of a problem along with a copious log, and we have to dive in and make sense of a very complex situation. The application has hundreds of settings and options, making the potential code paths practically infinite. Maybe we have a copy of the customer’s database with their precise settings, or maybe we just know they have one specific setting inside the problem feature set. Maybe we can reproduce the problem. Maybe we can’t. It’s almost always vague and confusing.
Here’s where agents can really shine. Normally, a huge log is a problem. What mere mortal can sift through the thousands of entries that our powerful logging tools can produce? No human, but Claude Code will happily gobble up all of that data and quickly zero in on the problem. And by quickly, I mean frequently in a matter of minutes or seconds.
The next choke point
Or as Spiros Xanthos, the CEO of Resolve AI — producer of a tool specifically designed for applying agents to production problems — says, “The real opportunity is purpose-built AI that can carry more of the cognitive load of production… If we get that right, developers spend less time manually piecing together what happened and more time building and improving the systems they own.”
I’ll venture to say that dealing with production problems is more challenging than writing code, both for humans and agents. But agents have huge contexts and infinite patience, and we humans do not. Code is a bounded system, and agents can train on an enormous body of data illustrating a vast array of coding techniques. Production? As Xanthos put it to me, it’s effectively unbounded, enormously varying, and a living, breathing system different for every product and every company.
We should always focus on the tightest bottleneck, and then move on to the next one. We’ve solved the coding choke point, and now we need to turn our attention to the next one — fixing production issues. That’s a non-trivial part of a developer’s job, and we shouldn’t be surprised that AI agents can do it better than we can.
Port has rolled out Port AI Builder, a vibe coding platform designed for software development and platform engineering teams.
Announced July 14, Port AI Builder lets teams create and run agentic workflows in natural language, with built-in human-in-the-loop review and approval. The platform lets organizations apply AI agents across the SDLC (software development life cycle), drawing on domain skills spanning site reliability engineering, devops, architecture, security, AI governance, data modeling, and UX, while maintaining governance and visibility, the company said. Port AI Builder is available through free and paid subscriptions.
Port AI Builder works on top of Port’s Agentic SDLC Platform, which provides the context lake, workflow orchestration, agent management, and governance that enterprises need to operationalize AI SDLC, according to the company. The new AI Builder lets teams build production-grade workflows in minutes, for use cases such as autonomous resolution, AI cost management, and engineering performance tracking, without losing control, Port said.
For decades, software engineering has been plagued by “productivity theater.” Every few years, the industry aligns around a new vanity metric — usually one that latches onto whatever technology happens to be in vogue at the time. For a discipline rooted in creativity and problem-solving, this is a poor way to demonstrate progress. Yet, we find ourselves in this position once again. The pattern is often the same: reach for something we can easily count, and in doing so, lose sight of what we are actually trying to achieve.
Quantity over quality: the wrong measurement, every time
I recall when I was coming up as a software engineer in the 1990s, a small number of companies took up the practice of paying their engineers by each line of code. This may have been productivity theater at its worst, leading to negative incentives, inefficient processes, and just generally bad engineering. Developers were rewarded for writing far more code than the problems they were facing required — classic “quantity over quality” — and the result was bloated, brittle codebases that were all but impossible to maintain. The goal — to create reliable software that solved real user problems — got buried under the incentive to produce.
Then in the 2000s, the rise of Agile brought us story points, an abstract way to estimate task complexity, effort, and risk relative to other work. Rather than answering “How long will this take?,” story points were meant to answer, “How big is this compared to what we’ve done before?” This approach sounds good in theory, but in practice, some development teams learned to game the system by inflating estimates, over-engineering solutions to look productive, and losing sight of whether the work they produced actually created value. Once again, the metric became the goal, and the actual goal — delivering outcomes that mattered to the business — became secondary.
Every one of these metrics failed for the same reason: they measured effort instead of value.
Quantity in the age of AI
Today, “tokenmaxxing,” a trend in which developers and teams optimize for consuming as many AI model tokens as possible, treats raw consumption as an equivalent for output. As I see it, this is the latest flawed productivity metric to make its way into the world of software engineering. Tokenmaxxing is nothing more than another vanity metric, and is just as useless as using “lines of code” or inflated “story points” as a benchmark.
Tokenmaxxing is the result of a few different behaviors, including:
Prompt flooding: stuffing massive codebases, documentation, and context into every prompt, burning tokens on context the model doesn’t actually need.
Agent swarms: running multiple AI agents in parallel to maximize code output, regardless of whether the work is coordinated or coherent.
Background loops: keeping AI sessions or agents running continuously in the background, racking up token spend without clear ownership of what is being produced — or why.
Now, it is no secret that AI is reshaping how software is developed, and these behaviors are the result of that reshaping. Providing AI with codebases, running multiple agents at once, and even relying on coding assistants for help all have their uses. But when we lose control of the changes we are making and why we are making them, we find ourselves facing a new version of the same old problem: measuring engineering productivity with the wrong metrics.
A more useful question to ask isn’t, “How many tokens did we spend?” but rather, “What problem did we actually solve, and for whom?”
Spending resources without goals
Yes, AI is giving software engineers the ability to do more with less, to move quickly, and to experiment in ways that were previously out of reach. But leaning on AI to perform productivity, rather than deliver it, is a trap that will cost us in code quality, team capability, and business credibility.
As a CTO, I am all for experimenting with AI. I want to use it to make our programs better, stronger, and future-proof. What I don’t want is for it to drive us toward excess while leaving us with little to show for it.
The test I keep coming back to is simple: does this AI-generated output help us ship something that matters? Does it reduce friction for a user, close a gap in a workflow, or improve reliability for a customer? If the answer isn’t clear, then we are spending resources — both human and computational — without a defined goal. And that is not engineering. That is activity.
Spec-driven development: where value gets defined
It is time to adopt newer approaches like spec-driven development, a method where engineers write detailed specifications first and AI generates code against them. Rather than relying on prompt flooding and agent swarms and hoping AI produces the best result, we need to shift toward defining requirements, reviewing AI-generated output, and orchestrating systems with intent.
But spec-driven development is more than a methodology. It is the place where engineering intent and business value get defined together. The spec is where you answer, “Why does this matter, and what problem are we solving?” before a single token gets spent.
Software engineers have long taken pride in writing elegant code, and I would hate to see AI cheapen that pride rather than elevate it. In an AI-first world, the craft shouldn’t disappear; it should simply move upstream. The spec is where elegance lives now, and it deserves the same attention to detail we once reserved for the code itself.
At its core, software engineering is about defining, analyzing, and resolving technical challenges. If we are willingly giving all of that up to AI, we will lose the integrity of our discipline and the ability to prove our value. Using the maximum number of tokens to produce code isn’t impressive. Using a well-crafted, intentional prompt to solve a specific problem? That’s the work worth celebrating.
Stop performing productivity and start delivering it
We are at an inflection point. Many organizations are defaulting to activity-based metrics, measuring how much AI is being used rather than whether it is improving delivery, product quality, or business outcomes.
The question worth asking is not, “How much AI did we use this sprint?” It is “What value did we deliver for our users, our team, or our business?” Was it the ability to resolve a critical bug more quickly? Reduced cycle time on a high-value feature? A customer workflow that now takes minutes instead of hours? Those are outcomes. Those are the things worth measuring.
AI can help us deliver meaningful outcomes faster, but only if we use it with the same rigor and intent we expect from every other engineering or business decision. Don’t let it become another form of productivity theater. The most successful engineering organizations in the age of AI won’t be the ones that consumed the most tokens, they’ll be the organizations that never lost sight of why they were building in the first place.
—
New Tech Forum provides a venue for technology leaders—including vendors and other outside contributors—to explore and discuss emerging enterprise technology in unprecedented depth and breadth. The selection is subjective, based on our pick of the technologies we believe to be important and of greatest interest to InfoWorld readers. InfoWorld does not accept marketing collateral for publication and reserves the right to edit all contributed content. Send all inquiries to doug_dineley@foundryco.com.
OpenClaw’s announcement that it has become a nonprofit foundation is generating IT excitement because of the potential for governance and development consistency that the popular platform has thus far lacked. Still, some worry about the risks created by the move.
“Our ambition is for OpenClaw to be the Switzerland of AI. Neutral ground where every model and every lab can plug into the technology and collaborate on standards in the era of agents,” OpenClaw said in a post. “That work is already underway in Foundation-convened councils on agent identity, agent profiles, evals, and enterprise deployment.”
The statement, co-authored by OpenClaw creator Peter Steinberger, pointed out, “the great open source projects of our time — Linux, Apache, Mozilla — endure because a neutral steward stands behind them. That is the role we are taking on to keep OpenClaw MIT licensed, open, and independent so that everyone building on it can trust it will be here for the long term.”
But it reassured users that the original OpenClaw leadership is still in charge.
“Peter built this thing and Peter keeps making the calls, especially the technical ones. Since joining OpenAI earlier this year, he has continued to steward OpenClaw as an open and independent project, and OpenAI has made a commitment to keep it that way,” the post said. “The foundation is here to serve: good governance, stable funding, and paying the people who keep the claws alive.”
However, some analysts and consultants were skeptical about how much true independence Steinberger would have, given his salaried role with OpenAI.
Neutrality claim in question
“The Switzerland of AI neutrality claim collapses under its own announcement,” said Noah Kenney, principal consultant at Digital 520. “OpenAI runs a team [at OpenAI] called Claw Labs that Peter leads and OpenAI is a major donor to OpenClaw. The ‘neutral steward’s’ chief technical decision maker is employed by one of the competing labs it is supposed to be neutral with.” To OpenAI, he said, OpenClaw is closer to a tax-exempt nonprofit subsidiary than it is to a neutral ‘Switzerland of AI.’
He pointed out that, in addition, Microsoft is shipping the enterprise version of OpenClaw, and Nvidia is shipping the hardware bundle. “This is being called the Switzerland of AI, but Switzerland does not have its central bank run by France,” he observed.
Kenney said that what the new OpenClaw has actually built is “a shared dependency that several competitors fund, staff, and steer, wrapped in a nonprofit structure. Enterprise IT should understand that structure, because treating OpenClaw as neutral is a mistake,” adding that CIOs need to look at this development devoid of the emotional component.
“There is a strategic irony here that CIOs should sit with,” Kenney said. “If OpenClaw succeeds at becoming the universal agent substrate, then every model plugs into the same identity layer, the same profiles, and the same deployment plumbing. The thing every vendor is racing to own becomes a commodity that nobody owns.” He pointed out that, in the short term, that is genuinely good news for buyers because it means less lock-in and more portability.
“But,” he said, “when the connective tissue is free and natural, the only labs that benefit are the ones with the best models and the deepest distribution. Commoditize the layer below you and you compete on the layer where you are already strongest. The foundation is not a charity. It is the biggest players agreeing to stop fighting over the plumbing so they can fight over the water, and the enterprise is the one paying the water bill either way.”
Good news, bad news
Jason Andersen, principal analyst at Moor Insights & Strategy, liked the potential consistency that could emerge from the structural change, given the complexity of agent development today.
“We are seeing a lot of OpenClaw variants hit the market, such as those from Nvidia as well as competing products from cloud and SaaS vendors. A common base helps solidify the common parts,” Andersen noted. “That said, a common challenge is the sustainability of these open source foundations over time. In addition to releasing code, these foundations need funding to evolve and grow. And that funding needs to come from continued momentum to incentivize existing members to increase investment and recruit new members to join.”
Andersen stressed that IT buyers need to keep an eye on the roadmap for any OpenClaw variant they choose to deploy, “as that will directly impact the foundation, and the momentum of the foundation and common base. If the common base loses momentum, it can lead to forks, or just a loss of innovation. When that happens, members tend to back away, which puts customers in limbo.”
But not everyone sees the promised structure as entirely good for IT.
Ishraq Khan, CEO at coding productivity tool vendor Kodezi, said, “most CIOs do not want to bet their future entirely on a single model vendor. They want Claude for some workloads, GPT for others, open models for sensitive environments, and potentially internally fine-tuned systems for specific use cases. The problem is that every vendor currently brings its own identity system, tool interfaces, permissions model, and operational assumptions. That fragmentation does not scale.”
He said, “the risk if standards fail is straightforward: every vendor builds its own closed ecosystem, enterprises become locked into individual stacks, and security becomes dramatically harder. The opportunity if OpenClaw succeeds is equally significant: enterprises get portable agents, common identity standards, interoperable tooling, and a healthier competitive market around models rather than ecosystems.”
Will it remain a nonprofit?
However, said Justin Greis, CEO of consulting firm Acceligence, one of the key details that IT executives will want to keep in mind is that OpenAI also began as a nonprofit, but it was quickly seen as not adhering to nonprofit objectives.
“OpenAI’s transition from a nonprofit research organization into a more complex structure highlighted the challenge of maintaining mission alignment while scaling technology, capital, partnerships, and commercial operations,” Greis said. “OpenClaw has the opportunity to address some of those governance questions earlier by establishing clear principles around neutrality, transparency, and decision-making before the ecosystem becomes even larger and more valuable.”
He noted, “we have seen this pattern before with technologies like Linux and Kubernetes. The strongest open ecosystems succeeded because they created trusted foundations that enterprises could build upon. The technology was important, but the governance model that underpinned it was equally critical.”
Risks are ‘squarely in IT’s lap’
Consultant Brian Levine, executive director of FormerGov, echoed Greis’ concerns.
“CIOs shouldn’t assume that this nonprofit will always be a nonprofit, or confuse being a nonprofit with actually being neutral or unbiased,” he said. “The risks are squarely in IT’s lap: autonomous agents ‘with their own identity’ acting on a user’s behalf blow straight through traditional IAM assumptions. Issues, such as agent identity, auditability, secret handling. Identity boundaries have not yet been reliably solved. Until they are, enterprises should treat OpenClaw agents like privileged service accounts, not like a browser plugin.”
Independent cybersecurity and risk advisor Steven Eric Fisher pointed to another IT exposure that might come from this OpenClaw transition: Cost.
“OpenClaw currently has a very high token burn rate in usage, which presents a significant cost consideration for large-scale enterprise adoption,” he said. “The skills marketplace introduces a new supply chain threat that enterprises will need to manage. Threat management, and specifically handling external marketplace elements, can be highly challenging for open-source operations. Ultimately, at scale, enterprise adoption could become a difficult balancing act between managing high operational costs and securing an expanded security surface.”
OpenClaw’s announcement that it has become a nonprofit foundation is generating IT excitement because of the potential for governance and development consistency that the popular platform has thus far lacked. Still, some worry about the risks created by the move.
“Our ambition is for OpenClaw to be the Switzerland of AI. Neutral ground where every model and every lab can plug into the technology and collaborate on standards in the era of agents,” OpenClaw said in a post. “That work is already underway in Foundation-convened councils on agent identity, agent profiles, evals, and enterprise deployment.”
The statement, co-authored by OpenClaw creator Peter Steinberger, pointed out, “the great open source projects of our time — Linux, Apache, Mozilla — endure because a neutral steward stands behind them. That is the role we are taking on to keep OpenClaw MIT licensed, open, and independent so that everyone building on it can trust it will be here for the long term.”
But it reassured users that the original OpenClaw leadership is still in charge.
“Peter built this thing and Peter keeps making the calls, especially the technical ones. Since joining OpenAI earlier this year, he has continued to steward OpenClaw as an open and independent project, and OpenAI has made a commitment to keep it that way,” the post said. “The foundation is here to serve: good governance, stable funding, and paying the people who keep the claws alive.”
However, some analysts and consultants were skeptical about how much true independence Steinberger would have, given his salaried role with OpenAI.
Neutrality claim in question
“The Switzerland of AI neutrality claim collapses under its own announcement,” said Noah Kenney, principal consultant at Digital 520. “OpenAI runs a team [at OpenAI] called Claw Labs that Peter leads and OpenAI is a major donor to OpenClaw. The ‘neutral steward’s’ chief technical decision maker is employed by one of the competing labs it is supposed to be neutral with.” To OpenAI, he said, OpenClaw is closer to a tax-exempt nonprofit subsidiary than it is to a neutral ‘Switzerland of AI.’
He pointed out that, in addition, Microsoft is shipping the enterprise version of OpenClaw, and Nvidia is shipping the hardware bundle. “This is being called the Switzerland of AI, but Switzerland does not have its central bank run by France,” he observed.
Kenney said that what the new OpenClaw has actually built is “a shared dependency that several competitors fund, staff, and steer, wrapped in a nonprofit structure. Enterprise IT should understand that structure, because treating OpenClaw as neutral is a mistake,” adding that CIOs need to look at this development devoid of the emotional component.
“There is a strategic irony here that CIOs should sit with,” Kenney said. “If OpenClaw succeeds at becoming the universal agent substrate, then every model plugs into the same identity layer, the same profiles, and the same deployment plumbing. The thing every vendor is racing to own becomes a commodity that nobody owns.” He pointed out that, in the short term, that is genuinely good news for buyers because it means less lock-in and more portability.
“But,” he said, “when the connective tissue is free and natural, the only labs that benefit are the ones with the best models and the deepest distribution. Commoditize the layer below you and you compete on the layer where you are already strongest. The foundation is not a charity. It is the biggest players agreeing to stop fighting over the plumbing so they can fight over the water, and the enterprise is the one paying the water bill either way.”
Good news, bad news
Jason Andersen, principal analyst at Moor Insights & Strategy, liked the potential consistency that could emerge from the structural change, given the complexity of agent development today.
“We are seeing a lot of OpenClaw variants hit the market, such as those from Nvidia as well as competing products from cloud and SaaS vendors. A common base helps solidify the common parts,” Andersen noted. “That said, a common challenge is the sustainability of these open source foundations over time. In addition to releasing code, these foundations need funding to evolve and grow. And that funding needs to come from continued momentum to incentivize existing members to increase investment and recruit new members to join.”
Andersen stressed that IT buyers need to keep an eye on the roadmap for any OpenClaw variant they choose to deploy, “as that will directly impact the foundation, and the momentum of the foundation and common base. If the common base loses momentum, it can lead to forks, or just a loss of innovation. When that happens, members tend to back away, which puts customers in limbo.”
But not everyone sees the promised structure as entirely good for IT.
Ishraq Khan, CEO at coding productivity tool vendor Kodezi, said, “most CIOs do not want to bet their future entirely on a single model vendor. They want Claude for some workloads, GPT for others, open models for sensitive environments, and potentially internally fine-tuned systems for specific use cases. The problem is that every vendor currently brings its own identity system, tool interfaces, permissions model, and operational assumptions. That fragmentation does not scale.”
He said, “the risk if standards fail is straightforward: every vendor builds its own closed ecosystem, enterprises become locked into individual stacks, and security becomes dramatically harder. The opportunity if OpenClaw succeeds is equally significant: enterprises get portable agents, common identity standards, interoperable tooling, and a healthier competitive market around models rather than ecosystems.”
Will it remain a nonprofit?
However, said Justin Greis, CEO of consulting firm Acceligence, one of the key details that IT executives will want to keep in mind is that OpenAI also began as a nonprofit, but it was quickly seen as not adhering to nonprofit objectives.
“OpenAI’s transition from a nonprofit research organization into a more complex structure highlighted the challenge of maintaining mission alignment while scaling technology, capital, partnerships, and commercial operations,” Greis said. “OpenClaw has the opportunity to address some of those governance questions earlier by establishing clear principles around neutrality, transparency, and decision-making before the ecosystem becomes even larger and more valuable.”
He noted, “we have seen this pattern before with technologies like Linux and Kubernetes. The strongest open ecosystems succeeded because they created trusted foundations that enterprises could build upon. The technology was important, but the governance model that underpinned it was equally critical.”
Risks are ‘squarely in IT’s lap’
Consultant Brian Levine, executive director of FormerGov, echoed Greis’ concerns.
“CIOs shouldn’t assume that this nonprofit will always be a nonprofit, or confuse being a nonprofit with actually being neutral or unbiased,” he said. “The risks are squarely in IT’s lap: autonomous agents ‘with their own identity’ acting on a user’s behalf blow straight through traditional IAM assumptions. Issues, such as agent identity, auditability, secret handling. Identity boundaries have not yet been reliably solved. Until they are, enterprises should treat OpenClaw agents like privileged service accounts, not like a browser plugin.”
Independent cybersecurity and risk advisor Steven Eric Fisher pointed to another IT exposure that might come from this OpenClaw transition: Cost.
“OpenClaw currently has a very high token burn rate in usage, which presents a significant cost consideration for large-scale enterprise adoption,” he said. “The skills marketplace introduces a new supply chain threat that enterprises will need to manage. Threat management, and specifically handling external marketplace elements, can be highly challenging for open-source operations. Ultimately, at scale, enterprise adoption could become a difficult balancing act between managing high operational costs and securing an expanded security surface.”
GitHub Copilot has introduced an interactive canvas that makes it easier to follow, review, and steer the actions of the upgrade agent as it modernizes .NET applications.
The GitHub Copilot upgrade agent assesses the application, generates a structured upgrade plan, creates implementation tasks, and executes the work, according to a July 9 bulletin from Microsoft. Now developers can follow that workflow in the GitHub Copilot app through an interactive upgrade canvas. The upgrade canvas provides a live view of the modernization workflow as it unfolds, Microsoft said.
This agent upgrade agent starts by assessing the .NET application and identifying what needs to change:
What version of .NET is this application targeting?
Which NuGet packages need to be updated?
Are there breaking API changes?
Which projects can be upgraded independently?
What should happen first?
From this point, the agent generates a structured upgrade plan and breaks the work into actionable implementation tasks. While the GitHub Copilot app provides the interactive upgrade canvas, GitHub Copilot upgrade also is available in the following tools:
Visual Studio – Built directly into Visual Studio. Right-click a solution or project in Solution Explorer and select Modernize to start a .NET upgrade.
Visual Studio Code – Install the GitHub Copilot upgrade extension, select the Upgrade agent from the agent picker dropdown menu, and prompt the agent to modernize your .NET application.
GitHub Copilot CLI – Install the GitHub Copilot upgrade plugin to assess, plan, and execute .NET upgrades directly from the terminal.
The feature was part of the new AI image generator, Muse Image, and allowed AI-generated images to be created using content from public Instagram accounts. By specifying a public account with an @mention, users could let the AI model use the person’s images as a reference without informing the account holder.
The launch quickly prompted concerns about privacy and the risk of abuse, with critics saying the feature could be used to create manipulated or offensive images based on other people’s public content.
Meta said its the goal was to give users a creative tool where they could decide for themselves whether their public content could be used. But it removed the feature after feedback that the feature did not meet expectations.
“For a very long time, spreadsheets forced you to learn spreadsheet syntax and spreadsheet ways of working,” said Eric Birnbaum, director of product management for Google Sheets. “But think about how many people need to use spreadsheets at work and don’t have the skill set to create the kinds of spreadsheets that can be really helpful for them.”
The addition of artificial intelligence can help handle that issue, he said, making the software more accessible to a wide range of office workers. “AI is unlocking the power of spreadsheets, taking on a lot of the difficult work that’s required to use them. That can be incredibly empowering for users,” said Birnbaum.
Gemini in Sheets is available at no extra cost to Google Workspace subscribers, though a paid AI Expanded Access add-on – costing $30 per user each month – is required to remove certain usage limits.
Features that have rolled out in recent months include the ability for a Gemini agent in Sheets to carry out multi-step actions such as formatting, analysis and data entry, and, more recently, the ability to create entire spreadsheets from a single prompt. A Fill with Gemini feature builds on the existing AI function, enabling users to automatically populate selected cells by detecting intent from information within a spreadsheet as well as from the web.
Another feature, Sheets Canvas (currently available in alpha), lets users generate interactive apps that update in real-time based on changes to spreadsheet data. This could be a kanban board for a sales pipeline, for instance, or an analytics dashboard that uses Sheets as its back-end data source.
Google claims users already see a range of benefits from Gemini in Sheets. According to an August 2025 survey of 200 Sheets users conducted by the company, the majority of knowledge workers (89%) said AI features in Sheets save them at least an hour a week, and 88% believe AI features have made them more confident in their data analysis skills.
The most popular AI use cases include creating spreadsheets and charts, analyzing data, and fixing broken formulas. How widely these features are actually used is unclear; Google declined to provide weekly usage statistics for Gemini in Sheets.
As the company embeds Gemini deeper into Sheets, questions remain about just how much businesses can trust AI tools to handle important business data, as well as what increased automation means for those who spend much of their day wrangling data.
Eric Birnbaum, director of product management for Google Sheets.
Google
Computerworld recently talked with Birnbaum about the potential benefits and challenges of the latest evolution of spreadsheet software. This interview has been condensed and edited for clarity.
As businesses become more focused on seeing value from AI investments, what measurable benefits are customers seeing from Gemini in Sheets – for example, time saved or other forms of return on investment? “Spreadsheets remain one of the universal languages for businesses, and we don’t anticipate that’s going to change anytime soon. But by bringing AI into the product where people work, we think it can significantly reduce the technical tax of data: the time spent understanding it, sourcing it, analyzing it, visualizing it.
“Historically, data professionals spent — let’s estimate it at 80% of their time — on the mechanical groundwork, data cleaning, crafting formulas, formatting, troubleshooting, and maybe only 20% of their time on actual strategic decision-making.
“We think that by offloading the manual, time-consuming, error-prone data work to Gemini, we can flip that ratio a bit, so humans can focus on the things that really matter: the high-value questions to ask and the judgment calls and decisions that get made from them. We’re starting to see evidence of that in our own user base and customer base.
“The second point to make is that AI can democratize data analysis to some extent, and make it available to many more users. For so many years, the spreadsheet was a gatekeeper; if you couldn’t speak the rigid language of spreadsheet formulas, you couldn’t extract the value from data. But by introducing these natural language interfaces, AI is separating the analytical capability from the technical literacy.
“If you can ask the right questions, or describe what you want in plain language, Gemini and Sheets can help you achieve your goals in a spreadsheet, even if you have minimal spreadsheet skills yourself.
“What we’ve seen so far from users and customers is that it’s incredibly empowering for people who might have been scared off by data analysis or dreaded opening spreadsheets in the past. You don’t need to go and wait for a data analyst to help you; you can go and do this work yourself in a spreadsheet.”
The flip side is, how confident can businesses be if more junior employees can take on higher-level analysis tasks by relying on AI? Given the propensity for AI models to hallucinate, to what degree can businesses trust that these tools won’t introduce errors into important business data? “It’s something we spent a ton of time thinking about. We’ve gone to great lengths to build these AI tools to collaborate with you, to show their work, explain what they did, and make sure that you can take over where they leave off.
“Our Sheets agent, for example, lays out a really explicit, transparent plan for you to review and approve before any data manipulation happens. It’s designed to do that in plain natural language in a way that the average user could understand, and then it gives you back that final summary, so you know exactly what it did and where it did it.
“The other thing is that the model is great at explaining things. If you inherit a spreadsheet that has some complex formula that you don’t understand, for example, the model does an amazing job of explaining how it works and what it’s doing.
“In many ways AI is not only making these features more accessible, but helping users feel more confident in the output. Human error is an inherent risk in manual data management, with or without AI. One misplaced comma or broken cell reference can completely corrupt an entire financial model, and it can be completely undetected.
“Our approach with AI in Sheets is to create this deliberate verification loop. You now have another spreadsheet expert working along with you, reducing the likelihood of these mistakes.”
Even if humans produce errors too, does it ultimately come down to accountability when AI is involved? “Our point of view here is that AI should be partnering with the knowledge worker who’s doing the work here. And everything that we’ve built is designed to be that partner. You might be able to offload tasks to the model, but we’re citing sources, we’re providing plans and explanations. We’re ultimately relying on the user to do that final verification.
“We spend a humongous amount of time focused on quality. We know that for AI to be useful in spreadsheets, it has to be reliable. When we launched Sheets Gemini Agent, for example, we were really proud that we set a state-of-the-art benchmark on the full SpreadsheetBench data set, which at the time exceeded competitors and near-human expert ability.
“But we know quality is never ‘good enough’ or done. We’re constantly working to improve quality for our users and customers, and for the use cases where they’re relying on AI most.”
What potential do you see for more agentic functionality in Gemini Sheets — for example, bringing in data from other sources, creating recurring reports, or taking more actions independently? “We’re listening closely to customers and users and building what they’re telling us they need. The agent is already capable of doing very complex multistep workflows, and we see users discover that the agent is extremely capable of doing end-to-end spreadsheet tasks. In terms of connectors, in an alpha we have connections available to HubSpot, Salesforce, and Mailchimp. We hope to expand that over time.
“There’s no path to have AI replacing analysts. I think AI is giving analysts more time back to actually do the more valuable parts of their job. Analysts that I work with are way more productive and impactful than they ever were before, because they can push that uninteresting spreadsheet grunt work off to the model, freeing up time for more interesting and impactful work.
“A great example: the visualizations I’m getting back from analysts nowadays are canvases instead of static charts that I can explore myself. It’s way more informative and useful than what I was accustomed to before.”
Looking ahead, do you expect a larger share of spreadsheet work to be carried out by agents, with humans setting goals and reviewing results? What will be the biggest change in how people use spreadsheets with AI? “The tasks are likely to stay similar and the use cases for spreadsheets are likely to continue to be relevant. The biggest change will be the ability to push the uninteresting spreadsheet grunt work off to a model and free up time for the user to do things that are more impactful, more interesting, more meaningful to the business.
‘Spreadsheets have historically been these like static containers where data goes to rest. I think AI can turn spreadsheets into these dynamic, localized software applications. And I do actually think this sort of changes the game for how people might use spreadsheets moving forward.
“It’s not just a passive grid full of numbers; spreadsheets are evolving to become these live, long-lived, collaborative applications. Employees can build these on the fly, like a basic CRM or supply chain dashboard in seconds. This is an area of investment for us moving forward, and we’re really excited to see how this evolves.”
AI assistants and agents, such as ChatGPT or Claude, might be able to analyze spreadsheet files and business data without users working inside a spreadsheet application. What do you think will keep Sheets central to the workflow, rather than simply making it one part of a wider AI-driven process?“We’re in constant touch with customers and users; it’s clear work is still happening in spreadsheets. I think spreadsheets remain incredibly popular tools. If we can bring the AI capabilities users need directly into the product where they already are, we’ll transform the way they work.
“I think we can be the front-end for some of this great AI innovation and the products that users are accustomed to today. Everything we build is guided by user feedback: users are telling us right now they want AI to help them do their everyday or more complex tasks, and they’re starting in Sheets today.”
Apple surprised the tech industry after financial markets closed Friday with news the company has sued OpenAI, alleging theft of trade secrets for ChatGPT hardware. The lawsuit particularly targets some senior ex-Apple employees now working at OpenAI.
Apple’s suit names two former employees — Chang Liu and Tang Tan, former vice president for product design, iPhone and Apple Watch — as well as OpenAI and that company’s recently-acquired firm, io Products, alleging “trade secret misappropriation and breach of contract.”
For its part, OpenAI issued a brief statement in response to the litigation. “We have no interest in other companies’ trade secrets,” the company said. “We remain focused on building innovative technology that empowers people everywhere.”
The allegations against Tan
Some of the claims and allegations included in Apple’s lawsuit include:
That in the months before leaving Apple, Tan met with OpenAI or its collaborators and discussed meetings with a key Apple supplier.
He emailed himself information about suppliers and internal summaries.
When interviewing former Apple staffers for jobs, he used confidential information, such as internal project code names, to gain even more knowledge.
He asked candidates to bring actual parts from Apple to interviews to discuss — and a then-Apple employee screenshotted and downloaded files concerning a highly confidential Apple project before attending an OpenAI recruitment session.
Tan asked Apple employees to bring CAD/design artifacts to their interviews.
Tan allegedly instructed new hires on how to avoid scrutiny when leaving Apple, such as instructing them not to tell the company they had taken jobs at OpenAI.
The ‘so funny’ laptop bug
The lawsuit also claimed that after quitting Apple for OpenAI in January 2026, Chang Liu managed to keep or “otherwise acquire” an Apple-issued notebook which he used to access confidential data on the company’s private network while at OpenAI. “LOL, I found out I can access the [server], so funny,” Liu texted a friend still working at Apple.
The suit alleges that he made no effort to report the situation, which was a bug in the system he had uncovered. Apple eventually discovered the exfiltration was taking place and took steps to prevent it, but Liu allegedly downloaded more than 1,000 pages of data, including “confidential technical presentations, spreadsheets, PDFs, and written work product,” Apple said.
“Only OpenAI and Mr. Liu know all the ways they have been exploiting the trove of Apple confidential information he stole, and to the extent they have not concealed or destroyed the evidence of these misappropriations, it will be investigated thoroughly in discovery.”
Apple’s lawsuit also alleges Liu was simultaneously coaching a current Apple employee named Alyssa Peng on how to copy files from Apple workstations without triggering the security team, asking her to get specific confidential information and using Apple’s stolen data to help her get ready for an eventual OpenAI interview.
Why this could get bigger
There’s much in the litigation that it will garner serious international attention as it unfolds. Apple’s argues that a competitor with access to so much of its own proprietary information could “bypass years of independent research and development, skip the capital expenditure required to build genuine expertise, and bring products to market faster and at lower cost, harming the value of Apple’s investments.”
It’s not just the secrets behind actively-used processes Apple is protecting; the company is also asserting its rights to regain control of information it has assembled over time concerning processes and manufacturing attempts that have failed. That’s understandable – you can invest a lot of money in finding out what doesn’t work and knowing that is a trade secret in itself.
“OpenAI coaches candidates to prepare for their interviews by studying Apple’s confidential engineering documentation, internal presentations, and proprietary technical materials,” the litigation claims. “OpenAI then uses its insider Apple information to ask detailed questions to extract more: about Apple’s proprietary tools, vendor management processes, engineering methodologies, manufacturing workflows, and supplier relationships, for example.
“OpenAI has turned to trade secret misappropriation to free-ride off Apple’s decades of innovation,” Apple said. “This is the tip of the iceberg.”
The lawsuit also confirms that Apple often designs and customizes the specialized machinery used in its suppliers’ factories, and that trade secrets concerning those efforts have been grabbed. The suit notes that OpenAI works with established Apple suppliers Foxconn, Luxshare, and Goertek on its own hardware.
The lawsuit points out that OpenAI now employs more than 400 Apple engineers and executives (including the company’s former Vision Pro Vice President), suggesting its entire approach to hardware recruitment is based on extracting Apple’s proprietary knowledge from potential hires.
“Apple lacks visibility into what’s been happening behind closed doors at OpenAI, where such misconduct is normalized and exemplified by leadership,” the lawsuit argues. “This much is clear, however: at every level, from members of its Technical Staff to its Chief Hardware Officer, and in coordination with business partners, OpenAI has been stealing Apple’s trade secrets and confidential information. As a natural result, OpenAI’s nascent hardware business now rests on the shakiest of foundations, rotten to its core by its illegal reliance on misappropriated trade secrets.”
With hundreds of billions in AI investment already racked up for 2026, it’s important to remember the entire industry is currently built on a mountain of debt — and much of this borrowed money is being spent on data center capacity. That’s true, even though consumers would probably rather have a cheap Mac than spend money on an AI subscription service.
All this debt is being amassed because a small number of people at a very small number of firms have decided to make huge investments in the tech, which at present requires huge quantities of energy, memory and data center capacity to run.
But it won’t always be this way.
A mountain of debt, but we’re short of memory
Look, the industry as it is now just doesn’t seem sustainable. Trillions are being spent and memory vendors are shifting capacity to make the high-value, high-bandwidth memory these server farms require — at the expense of traditional consumer electronic suppliers.
The rapid rollout just creates AI tech will need to be replaced, likely at greater cost, in a few years’ time. In a nutshell, the industry is spending trillions to make billions; Sequoia’s David Cahn estimates the AI revenue gap between infrastructure expenditure and the revenue to justify it has already fallen $600 billion a year short.
At some point, the VC money will run dry, after which it is inevitable deployment will slow and demand for all the components — including memory used in these large language model (LLM) data centers will fall. Some analysts think capex growth in the sector could halt by mid-2027.
At that point, memory vendors will have expensive production facilities and extensive defaults on their order books. If the 2027 prediction is true, those vendors will feel this impact in the form of reduced forward orders by the end of 2026.
The problem is that the investments have become so vast that any slowdown will have consequential effects across all sections of the economy.
After the gold rush
Almost certainly, the technology will continue to improve, and the problems we’re looking to solve today might no longer be challenges once fresh innovation strikes. So, what happens next?
Let’s think about memory, the biggest pain point at the moment and where we will (hopefully) find future innovation. At present, some of the largest LLMs sit inside data centers supported by vast quantities of memory. These machines are built to handle really complex tasks, but most of the time are used to search the web, deliver writing assistance and summarize documents. Those frequently-transacted tasks barely stretch the capabilities of these services and Apple, and others have already figured out how to run such tasks on device.
That’s the first obvious space in which to innovate – to invest in 1-bit data LLM systems to miniaturize and distill models so they actually run on the device you’re using, rather than relying on all those remote servers.
The Apple shopping list
Apple’s interest in 1-bit data LLM pioneer PrismML speaks volumes about where the iPhone maker sees LLM development going, as did its acquisitions of Kuzu Inc., WhyLabs Inc, Pointable Inc., and Datakalab Inc. in recent years.
The beauty of PrismML’s tech is what it can do. It was recently used to compress Alibaba’s huge 27-billion-parameter Qwen 3.6 model from 54GB down to under 4GB, running with all 27 billion parameters active simultaneously — all without sacrificing benchmark performance.
The kicker? It managed to run that advanced, sophisticated AI model on an iPhone 17 Pro. My take? Just as music used to be captured on reel-to-reel tape and is now digitized and in the air, AI will move from the data center to the device, possibly faster than people expect.
Apple has three pillars for AI: On-device for most of what you need, on Private Cloud Compute servers for most of the rest, or via third-party server-based systems for the most demanding tasks. That’s a blueprint for how the industry will evolve as technologies represented by PrismML tend toward bringing more of that intelligence to the device. Over time, those local tasks will become more sophisticated, eroding the available market for today’s heavily-indebted AI incumbents.
Emerging priorities such as the need for privacy, data sovereignty, and trusted cloud will also spur the emergence of a multipolar AI future in which no one vendor dominates, further complicating their journey to profitability. It’s a model that favors the kind of service-agnostic, edgeAI approach Apple has taken.
EdgeAI for the rest of us
In the end, I don’t think there will be a need for much of the AI data center capacity now being built, because Apple and others will figure out how to use data minimization to transact sophisticated AI tasks on the device. For the most part, EdgeAI will deliver the consumer AI experience, while data centers cater to more sophisticated use. One day, after this gold rush has run its course, we’ll peer outside of our basements to see which of today’s AI firms actually are the chosen ones.
They may not be the ones you expect.
You can follow me on social media! Join me on BlueSky, LinkedIn, Mastodon, and subscribe to the human-curated daily Apple news briefing at The Core.
Login
