Eco-friendly mechanochemical route yields antioxidant thiazolyl-hydrazone phenolic compounds

11 September 2026 at 11:43

In a finding that could reshape how medicinal chemists produce antioxidant compounds, a team of Italian researchers has developed a rapid, solvent-minimized ball-milling method for synthesizing thiazolyl–hydrazone derivatives — molecules that in laboratory tests proved to be more potent radical scavengers than vitamin C. The work, published in Results in Chemistry, combines green chemistry principles with computational modeling to deliver a class of compounds with significant promise for combating oxidative stress, the destructive process implicated in aging, cancer, and chronic disease.

The research, led by Roberto Scipione, Sebastiano Masuri, Ester Sedda, Tiziana Pivetta, Andrea Porcheddu, Maria Grazia Cabiddu, and Andrea Citarella at the University of Cagliari, tackles a long-standing challenge in synthetic chemistry: how to produce polyphenolic thiazolyl–hydrazones efficiently without drowning the process in solvents and energy. Traditional approaches require hours of refluxing in hot ethanol, consuming large volumes of solvent and generating substantial waste. The new protocol accomplishes the same transformation in two hours at room temperature using less than half a milliliter of ethanol.

The molecules at the heart of the study belong to a family of organic compounds that chemists prize for their antioxidant potential. Thiazole-containing heterocycles — five-membered rings containing both sulfur and nitrogen — are known to facilitate the decomposition of hydroperoxides and to chelate metal ions that would otherwise catalyze damaging reactions. Hydrazone linkers add a further dimension of reactivity: their nitrogen-hydrogen bonds can donate hydrogen atoms to neutralize free radicals, and the resulting radical species are stabilized through electron delocalization. When researchers attach catechol or guaiacol phenolic groups — structural motifs found in celebrated antioxidants like the polyphenols of olive oil and green tea — the molecules become even more effective at scavenging radicals, because the phenoxy radicals formed after hydrogen donation are exceptionally stable.

Oxidative stress arises when reactive oxygen and nitrogen species overwhelm the body’s endogenous antioxidant defenses. The resulting damage to lipids, proteins, and DNA is a hallmark of chronic inflammation, cancer progression, and aging. Designing small molecules that can intercept these reactive species has therefore been a major goal of medicinal chemistry for decades. The thiazolyl–hydrazone framework synthesized by the Cagliari team combines several radical-scavenging architectural elements in a single molecule, creating compounds whose potency in standardized assays exceeded that of ascorbic acid, the benchmark reference compound.

The synthetic breakthrough lies in mechanochemistry — the use of mechanical force, delivered by grinding balls in a laboratory mill, to drive chemical reactions without bulk solvent. The researchers used a Retsch MM500 mixer mill operating at 30 Hz with zirconia jars and balls, running the entire two-step sequence in a single jar without ever isolating an intermediate. In the first step, an aldehyde condenses with thiosemicarbazide to form a thiosemicarbazone; in the second, that intermediate cyclizes with a phenacyl bromide to close the thiazole ring. A catalytic amount of acetic acid and a tiny quantity of ethanol — a technique called liquid-assisted grinding, or LAG — proved essential: without any liquid additive, conversion was a dismal 10 percent, but adding ethanol at a ratio of one microliter per milligram of solid reagents pushed yields to around 85 percent.

The full telescoped protocol — both steps performed consecutively in the same jar — delivered the target compound in 80 percent yield, matching or beating the conventional two-step route that requires 12 hours of reflux at 80 °C in 25 mL of ethanol. When the team scaled the reaction up to 20 mmol using a 50 mL jar, the protocol held up remarkably well, affording a 71 percent yield with an even smaller amount of grinding liquid, simply reduced to a quarter of the original ratio. Product isolation required nothing more than adding ethanol and filtering — no column chromatography.

Green metrics quantified just how dramatic the improvement is. For the representative compound ES3, the mechanochemical route reduced solvent consumption by 99 percent, from 40.8 mL per gram of product to 0.38. The Process Mass Intensity, a measure of total material used per unit of product, plummeted from 35.3 to 2.05 — a seventeenfold improvement. The E-factor, which captures waste generation, fell from 34.3 to 1.05, roughly a thirty-threefold reduction. Reaction Mass Efficiency nearly doubled, rising from 32.3 to 57.1 percent, while overall yield climbed from 42 percent to 80 percent. These are not incremental gains; they represent a fundamental shift in the resource profile of the synthesis.

The antioxidant activity of the fifteen compounds in the series was assessed using two well-validated assays: DPPH and ABTS, both of which measure the ability of a molecule to neutralize stable free radicals. Every compound showed activity in the micromolar range, and several outperformed ascorbic acid in both assays. The standouts were those bearing the 3,4-dihydroxy (catechol) motif: compound ES26 recorded an EC50 of 8.4 μM in the DPPH assay and an extraordinary 2.59 μM in the ABTS assay, compared with 18.6 and 12.2 μM for ascorbic acid, respectively. Compounds ES27 and ES28 performed nearly as well, while several methoxy-substituted derivatives — ES9, ES16, and ES17 — also beat the vitamin C benchmark.

Density Functional Theory calculations provided mechanistic insight into why the catechol-bearing compounds are so effective. By computing bond dissociation enthalpies, ionization potentials, proton affinities, and related thermochemical descriptors, the team mapped the three canonical antioxidant pathways: hydrogen atom transfer (HAT), single-electron transfer followed by proton transfer (SET-PT), and sequential proton loss electron transfer (SPLET). The results indicate that in the gas phase, HAT is the thermodynamically preferred route, with the hydrazone NH group serving as the primary hydrogen donor in the simpler benzaldehyde-derived compounds. In vanillin-based molecules, hydrogen abstraction from the phenolic OH becomes competitive; in the catechol derivatives, abstraction from the 4-OH position is decisively favored, owing to the additional stabilization of the phenoxy radical by an intramolecular hydrogen bond with the adjacent hydroxyl group. In solution, the SPLET pathway gains relevance, as solvation stabilizes the anionic intermediates that form upon proton loss.

The structure–activity relationships that emerged are consistent with established antioxidant chemistry. Electron-releasing groups such as methoxy substituents on the thiazole ring enhance radical-scavenging activity, whereas electron-withdrawing bromine substituents diminish it — a trend that can be mitigated when a catechol group is present elsewhere in the molecule. The computational analysis of frontier molecular orbitals confirmed extensive π-conjugation across the molecules and showed how substituents tune the HOMO–LUMO gap in ways that correlate with observed reactivity. The authors note that the slightly reduced coplanarity in the dimethoxy-substituted ES26, evidenced by a dihedral angle of 172.9 degrees versus 176.95 degrees in its monomethoxy counterpart, subtly alters the electronic landscape of the frontier orbitals.

What distinguishes this study is the marriage of two agendas that are often pursued separately: the design of biologically active antioxidant scaffolds and the radical greening of the synthesis itself. Mechanochemistry has been gaining momentum across organic synthesis as a way to slash solvent use and energy input, but demonstrations that combine solvent-minimized one-pot telescoping with preparative scale-up and rigorous green-metrics accounting remain relatively rare. The Cagliari team has shown that a complex, multicomponent heterocyclic synthesis — involving condensation, cyclization, and the construction of a sulfur-nitrogen heterocycle — can be executed cleanly in a single milling jar without heating.

The implications extend beyond the laboratory bench. Antioxidant scaffolds like the thiazolyl–hydrazones synthesized here are candidates for further biological evaluation, and the authors indicate that future work will probe mechanistic aspects and explore preliminary biological evaluations to define the translational potential of these molecules. If the exceptional radical-scavenging activity observed in vitro carries through to cellular or in vivo contexts, the compounds could serve as leads for agents targeting oxidative-stress-related pathologies. And because the synthetic route is scalable, solvent-lean, and operationally simple, the barrier to producing gram quantities of these candidates for further testing is substantially lower than it would be with conventional methodology.

The study also underscores a broader lesson for the pharmaceutical and fine-chemical industries: green metrics such as PMI and E-factor are not abstract sustainability scores but concrete indicators of process quality — lower waste, lower cost, higher yield, and simpler workflows all at once. By demonstrating a sixfold reduction in reaction time, a near-total elimination of solvent, and improved yields in the same breath, the researchers have provided a compelling case study for how mechanochemistry can move from academic curiosity to practical synthetic platform.

As the search for effective antioxidant therapeutics continues, this work offers a template: design molecules around well-understood radical-stabilizing motifs, synthesize them with minimal environmental footprint, and validate their activity with rigorous in vitro and computational analysis. The thiazolyl–hydrazones that emerged from the zirconia jars in Cagliari — particularly the catechol-bearing champions ES26, ES27, and ES28 — now stand as both chemical achievements and potential starting points for the next generation of radical-scavenging agents.

Subject of Research: Green mechanochemical synthesis of thiazolyl–hydrazone phenolic derivatives and evaluation of their antioxidant activity

Article Title: Green mechanochemical synthesis of thiazolyl–hydrazone phenolic derivatives and evaluation of their antioxidant activity

Article References: Scipione, R., Masuri, S., Sedda, E., Pivetta, T., Porcheddu, A., Cabiddu, M. G., & Citarella, A. (2026). Green mechanochemical synthesis of thiazolyl–hydrazone phenolic derivatives and evaluation of their antioxidant activity. Results in Chemistry, 30, Article 103794. https://doi.org/10.1016/j.rechem.2026.103794

Image Credits: AI Generated

DOI: 10.1016/j.rechem.2026.103794

Keywords: mechanochemistry, ball milling, green chemistry, thiazolyl–hydrazone, antioxidant activity, DPPH assay, ABTS assay, catechol, liquid-assisted grinding, DFT calculations, E-factor, radical scavenging

Giant-nucleus cells emerge as early cancer markers in new study

Science Mag

By: Nathaniel Bowman

11 September 2026 at 11:21

For more than four decades, pathologists have glimpsed something unsettling under the microscope in the earliest stages of certain cancers: cells with nuclei far larger than they should be, swollen and distorted, lurking in tissue that otherwise appears on the verge of catastrophe. These so-called giant-nucleus cells, or karyomegalic cells, were first spotted in the 1980s in animal experiments involving iron-induced oxidative stress, yet their biological identity and their role in the genesis of cancer remained stubbornly obscure. Now, a team of researchers in Japan has provided the most detailed portrait of these cells to date, and their findings suggest that the enormous nuclei may be far more than a curiosity—they could be the founding population of early cancer, visible before any tumor exists, and potentially useful as a predictor of patient outcomes.

The study, led by Shinya Toyokuni, professor emeritus at Nagoya University Graduate School of Medicine, together with assistant professor Yingyi Kong and colleagues, was published in Redox Biology. The work addresses a deceptively simple question: when iron overload kills most cells in a tissue through oxidative damage, who survives, and what do the survivors become? Iron is an essential micronutrient, but in excess it is a potent carcinogen. Its chemistry is both the reason we need it and the reason it can harm us. Iron catalyzes the formation of hydroxyl radicals from hydrogen peroxide through the Fenton reaction, and these radicals shred DNA, oxidize lipids, and damage proteins. When the damage is severe enough, a form of programmed cell death called ferroptosis is triggered—an iron-dependent demise characterized by lipid peroxidation cascades that rupture the cell membrane. Ferroptosis has become one of the hottest topics in cancer biology precisely because it links iron metabolism, oxidative stress, and cell fate in a single mechanism.

Toyokuni’s team reasoned that if ferroptosis sweeps through iron-damaged tissue, the cells that escape it must be exceptional in some way, and those exceptions might be the seed of malignancy. To find out, they turned to a well-established animal model in which ferric nitrilotriacetate, a carcinogenic iron compound, is injected into rats, delivering a bolus of iron to the kidneys. The compound accumulates in the proximal tubules, where it catalyzes oxidative damage and, over months, produces renal cell carcinoma. In the experiments, the researchers collected kidney tissue before the injection and again one and three weeks afterward. Crucially, they also included a second experimental group: rats carrying a single mutated copy of the BRCA1 gene, the same gene that, when mutated in humans, dramatically raises the risk of breast and ovarian cancers and is associated with renal carcinoma in certain contexts.

The methodological heart of the study lies in its marriage of two technologies. The first is spatial transcriptomics, a technique that maps gene activity across individual cells within a tissue section while preserving the tissue’s original architecture—so researchers can see not only which genes are active but where in the tissue those transcriptional programs are operating. The second is computational nuclear morphometry, which quantifies the size, shape, and density of cell nuclei in the same samples. By overlaying the two datasets, the team could link the physical appearance of each cell—above all, the size of its nucleus—to its underlying transcriptional state. This integration is important because morphological atypia has long been the currency of diagnostic pathology, but connecting that atypia to molecular signatures at single-cell resolution is a recent capability.

What emerged within a week of iron administration was striking. Giant-nucleus cells appeared in the damaged kidneys, and these cells were not simply wounded cells awaiting death. They showed elevated expression of potent oncogenes, including Myc and Met, and they had acquired resistance to ferroptosis—the very cell death pathway that was eliminating their neighbors. In other words, the extreme oxidative environment had functioned as a brutal selective pressure, and the cells best adapted to survive it had activated a program that resembles, in miniature, the hallmark machinery of cancer. The enrichment of Myc, a transcription factor that drives cell growth and proliferation, and Met, a receptor tyrosine kinase involved in invasive growth, paints a picture of cells that are not merely enduring damage but reorganizing themselves around it.

The BRCA1 connection proved equally consequential. BRCA1 is best known as a guardian of the genome, a scaffold for the homologous recombination machinery that repairs double-strand DNA breaks. In the BRCA1-mutant rats, giant-nucleus cells emerged with distinctly different biological profiles from those in wild-type animals. The interpretation the researchers offer is that BRCA1 deficiency cripples DNA repair, so more of the damaged cells that would otherwise die or be repaired instead survive in their injured, genome-compromised state—becoming precancerous giant-nucleus cells. Electron microscopy and cellular respiration measurements reinforced this idea at the subcellular level: cells from BRCA1-mutant rats showed mitochondrial remodeling, with impaired iron handling and compromised respiratory function. Mitochondria, long appreciated as power plants, are also central regulators of ferroptosis and iron metabolism, and their disarray in the mutant animals may be both a consequence of BRCA1 loss and a driver of the abnormal iron biology that feeds karyomegaly.

Using their combined gene-expression and nuclear-shape data, the researchers went further and classified the giant-nucleus cells into six distinct types. One type possessed extremely large nuclei and gene signatures of stress-induced growth arrest—cells that appear to have locked themselves into a senescent-like state, consistent with profiles reported in earlier studies. Another type, found more frequently in BRCA1-mutant kidneys, showed heightened activity of cancer-related genes alongside elongated nuclei, a combination the authors interpret as a potential precancerous state. The taxonomy matters because it reframes giant-nucleus cells from a single enigmatic category into a spectrum, some members of which may be harmless or even protective while others carry genuine malignant potential.

One of the most provocative findings concerns the cells surrounding the giants. Even morphologically normal cells near giant-nucleus cells exhibited gene expression changes that appear to promote survival, suggesting that the karyomegalic cells remodel their local microenvironment in ways that extend beyond their own borders. In BRCA1-mutant kidneys, the team also documented a substantial increase in stromal cells—non-cancerous supportive cells including immune and connective tissue elements—around the damaged regions. The local tissue environment, the tumor niche in the language of modern oncology, was being reshaped long before any tumor could be called a tumor. This raises the possibility that a single conspicuous cell, easily identified by its nucleus under a conventional microscope, could serve as a sentinel for a whole neighborhood of incipient malignancy.

The leap from rat kidneys to human medicine came through two complementary analyses. First, the researchers mined TCGA-KIRC, a large public database of kidney cancer patients, asking whether tumors carrying gene-expression patterns resembling the cancer-prone giant-nucleus cells were associated with different clinical trajectories. They were: patients whose tumors shared the gene patterns of the cancer-prone giant-nucleus cells had shorter survival times, while those with patterns resembling the healthier giant-nucleus cell types had better outcomes. Not all giant-nucleus cells, in other words, are created equal, and the transcriptional fingerprints defined in the rat model appear to carry prognostic weight in human disease. Second, in a small pilot study, the team examined breast tissue from seven individuals with inherited BRCA1 mutations and fifteen without, observing abnormal nuclear features similar to those in the animal model in the mutation carriers. While the sample size is modest, the parallel hints that karyomegaly may be a cross-tissue marker of BRCA1-linked precancerous change.

The broader implications run in two directions. Clinically, the promise is that giant-nucleus cells, being easy to spot under the microscope, could function as early indicators of cancer—visible warnings in tissue that would otherwise be classified as merely damaged or inflamed. The TCGA survival data suggest that their gene-expression signatures might even stratify patients by risk once a tumor has formed. Scientifically, the study offers a concrete mechanism connecting an environmental exposome factor—iron-catalyzed oxidative stress—to a defined, transcriptionally characterized precancerous population, with ferroptosis resistance as the pivot. It also implies testable therapeutic hypotheses: if blocking mitochondrial remodeling or reversing ferroptosis resistance could prevent the persistence of the precancerous cells, early intervention might stop cancer before it starts. Toyokuni and his colleagues are careful to note that long-term studies will be needed to determine how giant-nucleus cells actually progress into tumor cells, whether such interventions can work, and whether nuclear shape combined with gene patterns can be validated as a diagnostic in larger patient cohorts. But the central message is already legible: some of the earliest chapters of cancer may be written in cells that have grown too large to be ignored.

Subject of Research: Animals

Article Title: Iron-catalyzed oxidative stress reveals an exposome-related ferroptosis-resistant karyomegalic niche in BRCA1-linked renal carcinogenesis.

Article References: Kong, Y., Shiraki, Y., Furuhashi, K., Maruyama, S., Imaoka, T., Enomoto, A., & Toyokuni, S. (2026). Iron-catalyzed oxidative stress reveals an exposome-related ferroptosis-resistant karyomegalic niche in BRCA1-linked renal carcinogenesis. Redox Biology, 95, Article 104293. https://doi.org/10.1016/j.redox.2026.104293

Image Credits: AI Generated

DOI: 10.1016/j.redox.2026.104293

Keywords: giant-nucleus cells, karyomegaly, ferroptosis, BRCA1, iron-catalyzed oxidative stress, spatial transcriptomics, renal carcinogenesis, mitochondrial remodeling, nuclear morphometry, precancerous cells, TCGA-KIRC, Nagoya University

Monterrey Amid Mountains

NASA Science

By: Michala Garrison

11 September 2026 at 05:01

The light-colored urban development of Monterrey, Mexico, fills most of the top half of the photo, and green parallel mountain ridges arc across the bottom of the frame.

August 26, 2026

The curving, parallel mountain ridges of the Sierra Madre Oriental are an eye-catching feature of northeastern Mexico’s landscape. The spot where these folds nestle up against Mexico’s second-largest metropolitan area captured the attention of an astronaut aboard the International Space Station, who took this photo on August 26, 2026.

Monterrey, the capital of the state of Nuevo León, is an industrial hub supporting heavy industries such as ironworks and steelworks, as well as manufacturing facilities for goods ranging from textiles to processed foods to glass and plastics. The metropolitan area is home to 5.3 million people, according to the 2020 census. And while the city has seen overall population growth since 1990, the number of people living within 5 kilometers (3 miles) of the city center has declined, researchers have found—a trajectory shared with many of Mexico’s metropolitan areas.

In Monterrey’s case, urban expansion runs up against some unforgiving terrain. Along the city’s southern edge, layers of limestone, deposited in the late Mesozoic era and then folded between about 80 and 50 million years ago, form the Sierra Madre Oriental. Over millions of years, weaker rock layers have eroded away, leaving behind the distinct ridgelines that bound Monterrey today.

The Río Santa Catarina carves through the mountains and onto the semiarid floodplain where the city lies. Because of the dry environment, the river carries little to no water for much of the time. But its channel is crucial for collecting runoff from summer rains and serves as an important natural area for plant and animal life within the city.

The river runs through Monterrey’s urban core and between several island-like protrusions of folded rock. One of these is the Sierra Las Mitras, a state nature reserve established in 2000. The mountain ridge rises approximately 1,500 meters (4,900 feet) over the city and provides a haven for wildlife. As conditions become cooler and wetter with higher elevations, vegetation turns from cacti and thorny shrubs on lower rocky slopes to oak and pine forests higher on the ridge. Cerro de la Silla (Mount Silla or Saddle Hill) is another prominent feature of the landscape, contrasting with the built environment.

Near the city’s border with the Sierra Madre Oriental sits Universidad de Monterrey, a host venue for the NASA Space Apps Challenge. This annual hackathon will take place in November 2026 in person and virtually at sites around the world. Participating teams use NASA and partner agency data to tackle challenges in fields such as software development, astrophysics, space exploration, and agriculture.

Astronaut photograph ISS075-E-70481 was acquired on August 26, 2026, with a Nikon Z9 digital camera using a focal length of 400 millimeters. It is provided by the ISS Crew Earth Observations Facility and the Earth Science and Remote Sensing Unit at NASA Johnson Space Center. The image was taken by a member of the Expedition 75 crew. The image has been cropped and enhanced to improve contrast, and lens artifacts have been removed. The International Space Station Program supports the laboratory as part of the ISS National Lab to help astronauts take pictures of Earth that will be of the greatest value to scientists and the public, and to make those images freely available on the internet. Additional images taken by astronauts and cosmonauts can be viewed at the NASA/JSC Gateway to Astronaut Photography of Earth. Story by Lindsey Doermann.

Downloads

August 26, 2026

JPEG (14.83 MB)

References & Resources

International Parks, Sierra de las Mitras. Accessed September 10, 2026.
NASA Earth Observatory (2020, November 27) A Sliver of Mexico’s “Mother Mountain Range.” Accessed September 10, 2026.
NASA Earth Observatory (2017, May 10) Mount Silla and Monterrey. Accessed September 10, 2026.
NASA Space Apps Challenge (2026) Monterrey. Accessed September 10, 2026.
Universidad Autónoma de Nuevo León, La importancia ambiental del Río Santa Catarina. Accessed September 10, 2026.

You may also be interested in:

Stay up-to-date with the latest content from NASA as we explore the universe and discover more about our home planet.

Great Balls of Fire

4 min read

An astronaut on the International Space Station was surprised to photograph a shower of light streaking through the darkness while…

Article

The World Cup From 250 Miles Up

4 min read

Over the years, astronauts aboard the International Space Station have photographed several of the cities hosting the 2026 FIFA World…

Article

Shaping the Emerald City

4 min read

The hills and basins of Seattle, Washington, have been sculpted by glaciers and modified by humans.

Article

Keep Exploring

Discover More from NASA Earth Science

Subscribe to Earth Observatory Newsletters

Subscribe to the Earth Observatory and get the Earth in your inbox.

Earth Observatory Image of the Day

NASA's Earth Observatory brings you the Earth, every day, with in-depth stories and stunning imagery.

Explore Earth Science

Earth Science Data

Open access to NASA’s archive of Earth science data

The post Monterrey Amid Mountains appeared first on NASA Science.

Python na análise de dados: ferramenta indispensável ou competência sobrevalorizada?

Tek Notícias

By: Fátima Caçador

10 September 2026 at 19:22

Usando a sua experiência enquanto formadora de análise de dados, Beatriz Duarte defende os benefícios do Python e a combinação entre conhecimento técnico, pensamento analítico e aplicação prática.

The post Python na análise de dados: ferramenta indispensável ou competência sobrevalorizada? appeared first on Tek Notícias.

New EAMA highlights showcase 30-year cohort generation research findings

Science Mag

By: Beatrice Stafford

10 September 2026 at 13:53

Thirty years after a small group of European professors of geriatric medicine first warned that the continent’s fastest-growing population was being cared for by physicians who had never been formally trained to care for them, the program they created to fix the problem is marking its anniversary with a scientific milestone of its own. A new special article published in European Geriatric Medicine documents the research output of the most recent graduating cohort of the European Academy for Medicine of Ageing, or EAMA, and in doing so offers a rare longitudinal measure of how a structured, pan-European leadership program in geriatric medicine translates into scholarly productivity, professional advancement and, ultimately, better care for older people.

The new paper, titled “‘Highlights of latest EAMA’: generation research outputs of the 30-year anniversary cohort,” was authored by Thomas Münzer of the Geriatric Clinic St. Gallen in Switzerland, Hannes B. Stähelin of Basel, and Barbara van Munster of the University of Groningen Medical School in the Netherlands, writing on behalf of the EAMA Board. It arrives at a moment when the demographic arithmetic of Europe has made the question of geriatric capacity more urgent than ever: the proportion of people over 65 continues to climb, multimorbidity, frailty and polypharmacy have become the daily reality of hospital medicine, and health systems across the continent still struggle to recruit and retain physicians trained specifically in the medicine of ageing. The article’s premise is simple but rarely examined with this degree of rigor — if you invest systematically in the education of mid-career geriatricians, what do you get back?

To understand the significance of the anniversary cohort’s research output, one has to go back to the early 1990s, when the group of European Professors in Medical Gerontology published a stark assessment in the journal Age and Ageing. Writing in 1994, Stähelin and colleagues documented that medical gerontology was taught unevenly, and in many countries not at all, in European medical schools. Geriatric medicine, despite being one of the few specialties whose patient population was guaranteed to expand every single year, was frequently a subordinate unit within internal medicine, staffed by physicians who had acquired their expertise on the job rather than through dedicated training. The professors’ conclusion was that Europe needed not merely more geriatricians but a new cadre of academic leaders — clinicians who could teach, publish, shape curricula and advocate for the specialty within their national health systems.

That diagnosis led, in 1995, to the founding of the European Academy for Medicine of Ageing, an intensive postgraduate program deliberately modeled on leadership academies rather than conventional continuing medical education. EAMA’s structure has remained remarkably stable across its three decades: a small, competitively selected cohort of physicians, typically already experienced clinicians, is brought together for a series of residential courses spread over roughly two years, interspersed with mentored projects that require participants to produce original scholarly work. The curriculum spans the full breadth of the medicine of ageing — from the biology of ageing and the physiology of frailty, through geriatric syndromes such as falls, delirium, incontinence and sarcopenia, to the organizational science of acute geriatric units, nursing home medicine, rehabilitation and end-of-life care. Crucially, faculty are drawn from across Europe, exposing participants to radically different national models of geriatric care and building a professional network that persists long after the formal courses end.

The “reloaded” reform of the program, described in a 2018 paper by Roller-Wirnsberger and colleagues in European Geriatric Medicine, formalized what three decades of experience had taught the academy’s faculty. That paper set out explicit competencies and standards for a European leadership program in geriatric medicine, defining what a graduate of EAMA should be able to do: not just diagnose and treat, but design services, lead multidisciplinary teams, mentor juniors, conduct and critique research, and represent the specialty at institutional and governmental levels. The competency framework aligned the academy with modern medical-education theory, in which learning objectives are specified in advance and assessed rather than assumed. It is against this framework that the anniversary cohort’s performance, as documented in the new article, can be read as an evaluation of the program itself.

The central finding of the new special article is that the latest cohort of EAMA graduates has generated a substantial body of research output — the “generation research outputs” of the title — spanning peer-reviewed publications, presentations and ongoing projects in geriatric medicine. Each participant in the program is expected to develop a scholarly project under faculty mentorship, moving from a research question through protocol design, data collection and analysis to presentation and, ideally, publication. For the 30-year anniversary cohort, the academy compiled and highlighted these outputs, creating what the authors describe as the “Highlights of latest EAMA” — a curated snapshot of what a single generation of emerging European geriatric leaders produced during and immediately after their training.

Why does this matter beyond the individual careers involved? The technical answer lies in how research capacity compounds. A single published paper changes one evidence base marginally; a cohort of trained researchers who continue publishing changes the trajectory of an entire specialty. Geriatric medicine has historically been under-represented in high-impact clinical research relative to the burden of disease it manages. Conditions such as delirium, frailty and multimorbidity were for decades excluded from major trials or treated as confounders rather than outcomes, precisely because the specialty lacked enough investigator-trained clinicians to insist on their inclusion. Programs like EAMA attack that bottleneck at its root: every graduate who learns to design a study, secure a mentor and navigate the publication pipeline becomes a node through which clinical questions from the bedside can flow into the literature and back.

The pedagogical machinery behind that outcome is worth examining, because it explains why the anniversary cohort’s productivity is treated by the authors as more than a curiosity. EAMA’s residential courses are built around case-based, small-group learning in which participants bring real clinical and organizational problems from their home institutions and work them through with international faculty. The mentored projects between courses impose deadlines and accountability — the two ingredients that, as anyone who has supervised doctoral students knows, most reliably convert good intentions into submitted manuscripts. The academy also functions as a peer-review incubator: draft papers are critiqued by senior European geriatricians before submission, giving participants exposure to the standards of international journals at a stage of their careers when feedback is most formative.

The anniversary article also carries a personal and institutional poignancy. Hannes B. Stähelin, one of the co-authors, is identified in the publication as deceased; he was among the founding professors whose 1994 analysis launched the academy, and his presence on the author line of a paper published in 2026 spans the full thirty-year arc of the program. The continuity from the GEPMG position paper through the 2002 retrospective review of the academy’s first years, published in Gerontology, through the 2018 competency framework, and now to this documentation of cohort research output, constitutes something rare in medical education: a thirty-year, multi-generation audit trail of a single intervention in specialty training, documented by the people who built it.

For health systems, the implications are concrete. Europe’s national models of geriatric care differ enormously — from the fully established departments of the United Kingdom and the Netherlands to countries where geriatrics is still fighting for recognition as an independent specialty. EAMA’s cohorts deliberately mix these national traditions, so that a geriatrician from a country without an acute geriatric unit model trains alongside colleagues who run such units daily. The research outputs of the anniversary cohort, the authors suggest, function as a form of technology transfer: projects conceived in one system are published in international journals and become adaptable blueprints for colleagues elsewhere. In an era when the European Union explicitly tracks health-workforce capacity, an academy that reliably produces clinicians who can lead services and generate evidence is a strategic asset, not a professional courtesy.

The article is also, implicitly, an argument about how educational interventions should be evaluated. Medical education research is notoriously dominated by satisfaction surveys and self-reported confidence measures, which correlate weakly with actual performance change. Research output — publications generated, projects completed, academic positions attained — is a harder, more objective endpoint, and the EAMA Board’s decision to compile and publish the anniversary cohort’s outputs signals a commitment to accountability by measurable results. It is a methodology other postgraduate leadership programs in medicine, from palliative care to rural health, could plausibly adopt: define the competencies, train against them, and then report what the trainees actually produced.

The publication timeline of the article itself — received in December 2025, revised in April 2026, accepted in June and published on 6 July 2026 in European Geriatric Medicine, the journal of the European Geriatric Medicine Society — places it squarely in the anniversary year it commemorates. As Europe’s population ages at a pace with no historical precedent, the question the founding professors asked in 1994 — who will teach the teachers, and who will lead the specialty? — has an answer that can now be pointed to in print: thirty years of EAMA cohorts, and a newest generation whose scholarly work is documented, highlighted and, the academy clearly hopes, replicated by the cohorts to come.

Subject of Research: Research outputs and scholarly productivity of the 30-year anniversary cohort of the European Academy for Medicine of Ageing (EAMA), a pan-European postgraduate leadership program in geriatric medicine

Article Title: “Highlights of latest EAMA”: generation research outputs of the 30-year anniversary cohort

Article References: Münzer, T., Stähelin, H. B., van Munster, B., & the European Academy for Medicine of Ageing (EAMA) Board (2026). “Highlights of latest EAMA”: generation research outputs of the 30-year anniversary cohort. European Geriatric Medicine. https://doi.org/10.1007/s41999-026-01536-5

Image Credits: AI Generated

DOI: 10.1007/s41999-026-01536-5

Keywords: European Academy for Medicine of Ageing, EAMA, geriatric medicine, medical education, research output, leadership program, ageing, Europe, geriatric care, anniversary cohort, lifelong learning

DCLK1 drives fibroblast activation and lung fibrosis via Smad3 binding

Science Mag

By: Nathaniel Bowman

10 September 2026 at 13:41

Idiopathic pulmonary fibrosis, a relentlessly progressive and fatal scarring disease of the lungs, has long frustrated clinicians with only two approved drugs that slow but cannot reverse its course. Now, a team of researchers in Taiwan has identified a surprising new player in the disease’s destructive machinery: doublecortin-like kinase 1, or DCLK1, a kinase better known for its role in cancer stem cells. Their study, published in the Journal of Biomedical Science, demonstrates that DCLK1 drives fibroblast activation in the lung and that blocking it—either genetically or with an oral inhibitor—slows fibrosis and preserves lung function in mice. The findings position DCLK1 as a promising therapeutic target in one of medicine’s most stubborn diseases.

The research, led by Lee-Yuan Lin, Wun-Hao Cheng, Chien-Huang Lin and Bing-Chang Chen of Taipei Medical University, together with collaborators at National Taiwan University Hospital and Universitas Gadjah Mada, began with a question that has lingered in the literature. Gene expression analyses had previously hinted that DCLK1 might be implicated in IPF, and earlier work reported that the kinase is markedly enhanced in the alveolar epithelial cells of patients with pulmonary fibrosis and in mice with bleomycin-induced fibrosis. But its function in fibroblast activation—the cellular engine of fibrosis—had remained unexplained.

To begin untangling that role, the team mined a publicly available single-cell RNA sequencing data set containing more than 312,000 cells from healthy lungs, IPF lungs, and lungs with chronic obstructive pulmonary disease. After stringent quality control filtering, they focused on roughly 240,000 cells from the control and IPF groups and mapped DCLK1 expression across the major lung cell lineages: epithelial, endothelial, lymphoid, myeloid, and stromal cells. The signal was unmistakable. DCLK1 was expressed across many cell populations, but the highest expression appeared in the stromal compartment, particularly in myofibroblasts—the specialized, contractile fibroblasts that churn out the extracellular matrix and stiffen lung tissue in IPF.

Human tissue told the same story. The researchers examined lung samples from eleven patients with IPF, obtained during lung transplantation or surgery, alongside fourteen nonfibrotic control samples. Histological staining with hematoxylin and eosin and Masson trichrome confirmed the expected picture of interstitial consolidation and collagen deposition in the IPF specimens. More importantly, immunohistochemistry revealed that both total DCLK1 and its phosphorylated, activated form were significantly upregulated in the fibrotic lung interstitium compared with control tissue. DCLK1, in other words, was not merely present in scarred lungs—it was switched on.

DCLK1 is an intriguing molecule to find at the center of a fibrotic process. It is a serine/threonine kinase belonging to the calcium/calmodulin-dependent kinase family, distinguished by two microtubule-binding domains. In oncology, DCLK1 overexpression has been documented in colorectal, pancreatic, gastric, renal, and hepatocellular carcinomas, where it correlates with poor prognosis, high recurrence, and metastasis, promoting tumor initiation and the epithelial–mesenchymal transition in cancer stem cells. It has also been linked to inflammatory signaling in SARS-CoV-2 infection, where inhibiting it blocks proinflammatory caspase-1/interleukin-1β signaling. The new study extends this portfolio of mischief to the fibrotic lung.

To test causality rather than mere correlation, the researchers turned to mice. They crossed animals carrying loxP sites flanking exon 3 of the DCLK1 locus with Pgk1-RFP-Cre/ERT2 transgenic mice, generating a system in which tamoxifen administration triggers global deletion of the Dclk1 gene. When these DCLK1 knockout mice and littermate controls were given intratracheal bleomycin, a chemotherapy drug used to induce pulmonary fibrosis in laboratory animals, the difference was striking. Knockout mice showed far less lung infiltration and collagen deposition, reduced numbers of cells staining positive for DCLK1 together with fibroblast surface protein and alpha-smooth muscle actin, and significantly lower levels of the profibrotic proteins fibronectin, α-SMA, and connective tissue growth factor, or CTGF.

The protective effect extended well beyond molecular markers. Using the FlexiVent FX system to measure lung mechanics, the team found that bleomycin treatment significantly reduced respiratory system compliance and pressure–volume performance and increased the normalized work of breathing in control mice—hallmarks of a stiffening, fibrotic lung. In the DCLK1 knockout animals, these functional deficits were substantially attenuated. In vivo microcomputed tomography, acquired at 34.75-micrometer voxel resolution, painted the morphological counterpart: bleomycin caused parenchymal destruction and volume loss in control lungs, while knockout mice retained lung structure and volume. Density histograms shifted toward fibrotic tissue in wild-type animals but stayed largely preserved in the knockouts, and peripheral blood neutrophil percentages, which climbed after bleomycin injury, were significantly reduced in the knockout group.

With the in vivo evidence established, the investigators dissected the molecular wiring in cultured normal human lung fibroblasts. Transforming growth factor beta, the master profibrotic cytokine, induced DCLK1 phosphorylation within 30 minutes and DCLK1 expression within 8 hours. Chromatin immunoprecipitation assays showed that after TGF-β stimulation, both Smad3—the canonical transcriptional mediator of TGF-β signaling—and the nuclear factor kappa B subunit p65 enriched at the DCLK1 promoter, indicating that TGF-β switches on DCLK1 through both Smad-dependent and noncanonical NF-κB pathways. Silencing DCLK1 with small interfering RNA then blunted the fibroblast’s profibrotic response, reducing TGF-β-induced expression of fibronectin, collagen 1A1, α-SMA, and CTGF.

The pathway, however, does not stop there. The team found that TGF-β also phosphorylated Akt, and that transfecting fibroblasts with a dominant-negative Akt construct reduced TGF-β-induced DCLK1 phosphorylation and the expression of fibronectin and CTGF—placing Akt upstream of DCLK1. Downstream, DCLK1 knockdown reduced Smad3 phosphorylation while DCLK1 overexpression enhanced it. Coimmunoprecipitation revealed a physical association between DCLK1 and Smad3, and immunofluorescence microscopy showed the two proteins colocalizing in the nucleus within 30 minutes of TGF-β treatment. A ChIP assay using a DCLK1 antibody pulled down DNA from the Smad3-binding region of the CTGF promoter, and pretreatment with the selective DCLK1 inhibitor DCLK1-IN-1 reduced TGF-β-induced CTGF expression. Together, the results sketch an Akt/DCLK1/Smad3 axis through which TGF-β drives the production of connective tissue growth factor, a potent profibrotic mediator.

The clinical payoff came in the pharmacological arm of the study. Mice receiving oral DCLK1-IN-1 at 10 milligrams per kilogram daily, beginning on day 7 after bleomycin instillation and continuing until day 28, showed visibly reduced fibrotic progression on histology, lower expression of fibronectin, α-SMA, and CTGF in lung tissue and lysates, preserved pulmonary compliance, improved pressure–volume loops, and reduced work of breathing. Serial microCT scans on days 7, 17, and 27 documented less severe fibrotic remodeling in the treated animals, with lung attenuation histograms showing a smaller fraction of dense, nonaerated voxels. Blood neutrophil counts were also lower. In short, a pill-like oral inhibitor measurably slowed the disease in a model that mirrors the human condition.

The implications are significant because the current therapeutic landscape for IPF is so thin. Only nintedanib, a tyrosine kinase inhibitor targeting growth factor receptors, and pirfenidone, which suppresses TGF-β production and activity, are approved by the US Food and Drug Administration, and both merely slow progression. The recent failure of pamrevlumab, a monoclonal antibody against CTGF, in the phase 3 ZEPHYRUS-1 trial suggested that targeting downstream fibrotic end products alone may be insufficient—pointing to the value of upstream signaling regulators such as DCLK1. Adding to the appeal, a DCLK1 inhibitor has already been shown to prevent inflammatory responses in acute lung injury, hinting at dual anti-fibrotic and anti-inflammatory potential.

The authors are careful to note limitations. DCLK1 is expressed across multiple lung cell types, and the tamoxifen-inducible knockout used here was global rather than cell-type-specific, so the contributions of fibroblasts versus epithelial, endothelial, and immune cells could not be fully separated. Human control tissue, as in most fibrosis studies, came from heterogeneous sources, including tumor-adjacent lung. The precise subcellular mechanics of the DCLK1–Smad3 interaction, whether DCLK1 inhibition can promote fibrosis resolution rather than merely slow progression, and the systemic safety of long-term DCLK1 blockade—all relevant given the kinase’s role in cancer and other processes—remain open questions.

Even so, the convergence of evidence is striking: DCLK1 is upregulated in human IPF, amplified in experimental fibrosis, mechanistically coupled to the TGF-β–Smad3 engine of fibroblast activation, and its suppression by both genetics and a small molecule preserves lung structure and function. What began as a cancer stem cell marker has now emerged, unexpectedly, as a central node in the scarring of the lung—offering fresh hope that a disease long considered irreversible may one day be stopped at its molecular source.

Subject of Research: Medicine

Article Title: DCLK1 drives fibroblast activation and lung fibrosis via Smad3 binding

Article References: Lin, L.-Y., Cheng, W.-H., Wu, Y.-C., Wen, H.-C., Hsu, H.-H., Liu, C.-H., Chen, J.-Y., Chang, W.-S., Wu, Y.-J., Wu, J.-S., Yuliani, F. S., Wang, C.-Y., Lin, C.-H., & Chen, B.-C. (2026). Doublecortin-like kinase 1 promotes fibroblast activation and fibrotic progression through Smad3 binding in idiopathic pulmonary fibrosis. Journal of Biomedical Science, 33(1), Article 52. https://doi.org/10.1186/s12929-026-01258-7

Image Credits: AI Generated

DOI: 10.1186/s12929-026-01258-7

Keywords: DCLK1 and cancer stem cell pathways in fibrosis, DCLK1 kinase in lung disease, fibroblast activation mechanisms in IPF, fibroblast-driven lung scarring, genetic suppression of DCLK1 in fibrosis models, lung fibroblast activation, lung fibrosis progression and molecular drivers, oral DCLK1 inhibitors for pulmonary fibrosis, pulmonary fibrosis, role of DCLK1 in lung tissue remodeling, Smad3 signaling pathway in fibrosis, therapeutic targeting of DCLK1

Why enterprises should start with on-site AI agents

InfoWorld

10 September 2026 at 11:00

When I talk to people about agents interacting with websites, the conversation almost always starts with perception: how does an agent “see” a page? Is it screenshots or the DOM or the accessibility tree? In my previous InfoWorld article, “Accessibility is the first-class interface for AI agents”, I argued that the accessibility tree is becoming an increasingly important interface between websites and AI agents. From there the conversation shifts to actuation, and to ways of making it less brittle, which let a site expose its own actions as callable tools instead of forcing an agent to guess at buttons and forms.

That conversation matters. But it leaves out a more consequential architectural question, one that determines who is accountable when an agent gets something wrong: where does the agent live?

There are three plausible homes for an agent that interacts with your business: on-site, in-browser and off-browser. Each one trades control for reach differently. Right now, much of the industry’s attention is focused on the two homes furthest from the business: browser extensions that co-browse alongside a user, and autonomous agents running in the cloud that talk to your APIs or drive a browser on their own.

I think that is the wrong place for most enterprises to start.

If you run a business with real compliance exposure, or a brand you cannot afford to have misrepresented, the pragmatic move is to build the agent that lives on your own site first and treat reach as a problem you solve later, on your own terms.

Reach without governance is a liability

The appeal of in-browser and off-browser agents is obvious. They promise discovery, scale and automation you do not have to build or maintain yourself. An off-browser agent can compare your prices against every competitor and keep working after your support team has gone home. A co-browsing extension can act inside a user’s already authenticated session without requiring any special integration on your side.

But an agent you do not control is still speaking for your business.

An off-browser agent talking to your APIs or scraping your DOM runs on a model you did not choose, follows guardrails you did not write and represents your brand in a conversation you cannot see. If it misquotes a price, misstates a policy or initiates a transaction that creates a customer dispute, the fact that the agent belonged to somebody else may not make the resulting problem disappear.

There is also a deeper security problem. Computer-use agents interact with untrusted content by design. Anthropic’s guidance for browser and computer-use agents notes that webpages and application interfaces can contain adversarial instructions and recommends scoped permissions, human confirmation for high-stakes actions, and monitoring of agent behavior. OWASP’s AI Agent Security Cheat Sheet goes further, identifying agent-specific risks including prompt injection, tool abuse and privilege escalation, data exfiltration and memory poisoning.

In-browser agents are a better deal, because a human is watching the same screen and can catch a mistake before it lands. But the model, the tone and the decision logic are still outside your hands. For or any other regulated industry, that is not a risk to take lightly just to pursue reach you have not yet learned how to govern.

What you actually gain by building at home

An on-site agent is less glamorous than the idea of your product being discovered by roaming agents across the open web. But it is the only home where the business keeps full governance.

You choose the model. You define the guardrails. You decide the tone, the escalation paths and exactly what the agent is and is not allowed to promise a customer. Because the agent has direct access to your systems, it does not need to reverse-engineer your checkout flow as an outside agent would. It can be given your actual domain knowledge and your actual APIs.

That matters because enterprises have solved this problem before. They did not expose databases directly to the internet; they introduced APIs. APIs were not just about integration. They created contracts, authentication, auditing and policy enforcement. On-site agents represent a similar evolution. Rather than forcing an external agent to infer behavior from a user interface, a business can expose governed capabilities through tools with explicit contracts.

WebMCP is a promising step in that direction. It makes it easier for a site to publish actions an agent can invoke, instead of forcing the agent to guess at the page. The important shift is not that the agent can click a button. It is that every action becomes observable, policy-governed and replayable.

The Chrome team’s WebMCP Sports demo is a good preview of what this can look like done well: an on-site assistant calling the site’s own WebMCP tools, with every call visible in the chat log. That is not a black box you are hoping behaves. It is an auditable interaction, with named tools and defined contracts.

The three homes

The cleanest way to think about this is as a spectrum.

On-site. The agent lives on your website. You ship it. You own the model, the tools, the tone, the policies and the telemetry. It has first-party access to your systems and your workflows. The trade-off is reach: it is coupled to your site, and it is not meant to carry the user’s journey across the web.
In-browser. The agent lives inside the user’s browser session, usually as a co-browsing extension or a browser-integrated assistant. It benefits from the user’s logged-in state and can act with the user present. That gives it real reach and a strong human safety net. But the business does not control the model or the guardrails, and the agent can still misrepresent the brand.
Off-browser. The agent runs elsewhere — in the cloud, on a server or on a user’s local machine and reaches your business by talking to APIs or driving a browser remotely. This is the widest-reach model, but also the least governed from the business’s point of view. It may compare competitors, cross workflows and act at scale, but you do not control the experience.

None of these homes is inherently bad. They solve different problems. But if your first priority is governance, accountability and brand integrity, on-site is the right starting point.

Start with the governed interface

For teams deciding where to put their first investment, three moves make the difference between a real interface and a chatbot glued to a homepage.

First, ship the on-site agent as the authoritative interface, not a fallback. It should have first-class access to your own APIs and domain knowledge — the same systems your support team and checkout flow already use — rather than being handed a scraped version of your own page.

Second, instrument every tool call it makes. If your agent calls a WebMCP tool to check inventory or apply a discount, log it the way you would log an API call from any other client. That log is your audit trail, your debugging tool and eventually your evidence when a regulator or customer asks what happened.

Third, design for human escalation from day one. Even a well-governed agent should not make every decision autonomously. Build explicit handoff paths to human operators for ambiguous or high-risk situations.

Taken together, those choices turn an on-site agent into a governed interface rather than a novelty feature. That distinction matters. The first is something an enterprise can safely operate; the second is something it will eventually have to defend.

When agents start talking to agents

There is another reason to build the on-site agent first: the three homes are unlikely to remain isolated.

Imagine that a user’s browser agent arrives at an airline website. It already knows the traveler’s preferences, calendar constraints, loyalty programs and perhaps their budget. The airline’s on-site agent knows something completely different: routes, seat availability, fare rules, upgrade eligibility and the airline’s operational policies.

Why should the visiting agent reverse-engineer all of that from a website?

A more natural architecture may be for the two agents to communicate.

This is no longer purely hypothetical. The Agent2Agent (A2A) protocol is explicitly designed to allow independent agents built by different vendors and frameworks to discover capabilities, exchange information, delegate work and coordinate tasks.

In that world, the user’s agent brings intent and user context. The site’s agent brings domain knowledge and governed execution.

And suddenly the on-site agent’s limited reach does not look quite so limiting. It does not need to travel across the web if other agents can come to it.

The incentive problem nobody is pricing in yet

There is a harder question underneath the technical one: who does the on-site agent actually work for?

A visiting agent, co-browsing or otherwise, is acting on the user’s behalf. An on-site agent, however well-built, is the business’s employee. And the moment the two negotiate a discount, an upgrade or a substitution, that exchange stops looking like a neutral handoff and starts looking like a sales conversation with an AI on each side of the table.

That tension is going to matter.

Businesses building on-site agents now should design their guardrails with that scrutiny in mind, rather than treating “the agent decided” as a shield. Control is worth having. It is also a responsibility. And building the agent that answers to you is the only way to exercise it.

The first agent should be the one you can govern

The history of enterprise software suggests a familiar pattern. Organizations adopt new interaction models by first bringing them under their own governance. Internal systems become APIs. APIs become platforms. Platforms become ecosystems.

AI agents will likely follow the same path.

That is why I think the first agent your business builds should not be the one that can reach the entire web. It should be the one whose behavior you can understand, measure and improve. Once that foundation exists, expanding outward becomes an architectural choice instead of a leap of faith.

Reach is tempting. But for regulated and brand-sensitive businesses, control has to come first.

Build your AI stack like you will need to replace it

InfoWorld

10 September 2026 at 10:00

Every AI feature you’re building is based on a price that isn’t real. Current AI services are heavily subsidized as model providers seek to expand their market share, and subsidies inevitably end. Just think back to the early days of Lyft and Uber, when a single ride often cost less than taking public transit. As the costs associated with building on cheap AI infrastructure slowly start to reflect reality, the architecture decisions you made to get to market quickly may be the ones hemming you in.

This isn’t unique to AI. IT leaders have always had to weigh speed-to-value against the cost of dependency. What’s different now is how fast those dependencies are being built, and how little visibility most teams have into them.

Across IT, scalability is contingent on portability. And architecting with portability in mind is the best way to minimize lock-in as AI pricing models evolve, giving you leverage in future contract negotiations, and a fighting chance of switching platforms if the costs of one vendor get too high.

The challenge is that AI providers are incentivized to make leaving as painful as possible, which they achieve through deep integrations and proprietary features. You can control some of this through architectural decisions. Other parts are structural dependencies you should at least be aware of going in.

Swapping platforms will never be painless, but it’s less painful than being locked in when your vendor decides it’s time to recoup its losses, as every AI provider eventually will.

The part you control

The foundation for portability is the abstraction layer, an internal service your applications call instead of hitting Claude, GPT, or Gemini directly. It uses a request format defined by you, and translates that to whatever the vendor behind it expects. If you need to change vendors later, it’s closer to a configuration change than a code rewrite.

You don’t need to build this from scratch; open-source gateways like LiteLLM or Portkey already handle the routing and interface. Scoped to what you actually need, it’s a small project that one or two engineers can complete in a few weeks.

From there, audit the parts of your codebase that call a vendor’s API directly and route them through your proxy layer one at a time (have fun with that). Skip the features you’re not using yet, like fine-tuning or batch APIs, and focus on the 80% case. Think: text in, text out, and maybe tool calls.

The parts you don’t control

The API call is the easy part to abstract. Other dependencies may be tougher to account for given the pressure to deploy quickly, but you should at least be aware of them.

Prompts get tuned to a specific model’s characteristics, and those instructions won’t necessarily make sense to another model. If you do switch vendors, expect to re-tune each prompt. Fine-tuning a model will box you in, unless you use an open-weight model that you operate internally.

Vectors generated by one provider’s model won’t be compatible with another’s, so switching means re-embedding your knowledge base. Tool calling and structured output conventions differ by vendor, so code written around one model’s schema will need rewriting for another. Caching and batching mechanics carry their own vendor-specific rules and cost implications that don’t map across providers.

In short, code portability is cheap. Behavioral portability is the real cost, and no gateway tool solves that for you.

Build with your eyes open

Most teams don’t realize how locked into a model they’re becoming until they’re forced to move, which is the worst time to find out. How many parts of your codebase import a vendor SDK directly, versus calling a shared internal function?

If you want a real answer instead of a guess, pick one workflow that isn’t business-critical and try porting it to a different vendor. See what breaks and how long it takes to fix. That will uncover your dependencies better than a checklist.

The bottom line is that you should treat AI providers like any other infrastructure vendor where concentration risk matters. Hold AI providers to the same standard you would apply to any IT or technology vendor. There will likely always be some level of dependency, but you should know where the hooks are and choose them deliberately. That way, you’re prepared when AI price hikes inevitably arrive.

—

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.

Your frontend observability has an accessibility blind spot

InfoWorld

10 September 2026 at 10:00

Frontend teams have become pretty good at monitoring what happens after an application reaches production. We track JavaScript errors, API failures, latency and Core Web Vitals. When something changes after a release, these signals help us determine whether the application is healthy.

But there is one gap I think frontend teams often overlook. An application can look completely healthy in those dashboards while an important workflow is unusable for someone navigating with a keyboard or assistive technology.

Imagine a checkout flow where a recent UI change makes the final button unreachable by keyboard. The page still loads. The API responds successfully. There are no new JavaScript exceptions, and performance metrics look normal. Every traditional signal says the application is working, yet some users can no longer complete the purchase.

That is why I have started thinking about accessibility as more than a compliance or pre-release testing concern. If an accessibility regression prevents someone from completing a critical workflow, that is also a production reliability problem.

Where traditional frontend observability falls short

Most frontend observability focuses on technical health. We want to know whether pages load quickly, requests succeed and deployments introduce errors. These are useful signals, but they mainly tell us whether the software is functioning as the system expects.

Accessibility failures are different because they often do not produce obvious technical errors. Removing an accessible name from an icon button will not necessarily throw a JavaScript exception. Replacing a semantic button with a clickable div might leave mouse interaction working normally. A focus-management regression inside a modal can make the interface extremely difficult to navigate while having no effect on page performance.

This creates a blind spot. Our monitoring can tell us that the button rendered and that clicking it successfully triggered an API request. It may not tell us that a keyboard user cannot reach that same button.

Automated accessibility checks in CI are an important first line of defense, and I use them as an early guardrail. But I would not treat a passing CI check as proof that the production experience is accessible. Production applications contain changing data, feature flags, third-party scripts and combinations of components that may not be represented in isolated tests.

Automated tools also have limits. They can identify many structural problems, but they cannot detect every accessibility or WCAG conformance failure. CI should therefore be an important checkpoint, not the end of accessibility verification.

Monitor accessible journeys, not just healthy pages

One way I think frontend teams can close this gap is by applying the same idea behind synthetic monitoring to accessibility.

Instead of checking only whether a page loads, we can test whether important user journeys remain usable. Can someone navigate the primary controls using only a keyboard? When a dialog opens, does focus move somewhere sensible? Can a form be completed without relying on a mouse? Do important controls still expose useful names and roles?

Consider the checkout example again. Traditional synthetic monitoring might verify that the page loads, products appear and the checkout API responds. An accessibility-aware journey could go further by navigating through the same flow with keyboard interactions and verifying that focus reaches the controls required to complete the purchase.

I would not try to automate every accessibility requirement this way. The goal is to identify a small number of critical workflows and ask a simple question: Can a user still complete the task?

For applications with a design system, I would extend the same thinking to shared components. A regression in a button, dialog, form field or navigation component can spread across many screens at once. Testing keyboard behavior, semantics, accessible names and focus expectations at the component level can catch problems closer to their source instead of discovering the same regression page by page.

This creates two useful layers of protection. Design-system checks help protect the building blocks, while synthetic journeys help verify that those building blocks still work together in real application flows.

Make accessibility part of release health

Frontend teams already look at signals after deployment to decide whether a release is healthy. We watch error rates, latency, failed requests and performance changes. I think accessibility signals can become part of that same conversation.

That does not mean creating another huge dashboard or turning every automated accessibility warning into a production incident. The useful approach is to focus on impact.

If a release introduces a minor issue in secondary content, the team may address it through the normal development process. If the same release prevents keyboard users from completing checkout, submitting a form or navigating a primary workflow, that deserves a different level of attention.

This is where accessibility-aware release health becomes valuable. Teams can define a small set of expectations around critical journeys and shared components, then watch for regressions just as they would watch other frontend reliability signals.

The need for this approach is becoming harder to ignore. The 2026 WebAIM Million analysis found an average of 56.1 detected accessibility errors per home page, a 10.1% increase from 2025. It also found that the average home page contained 1,437 elements, a 22.5% increase in a single year. As interfaces become larger and more dynamic, relying only on occasional accessibility audits becomes increasingly difficult to scale.

If users can’t complete the workflow, the system isn’t healthy

Frontend observability has given us increasingly sophisticated ways to understand what applications are doing in production. But our definition of application health should not stop at whether JavaScript executed, APIs responded or pages rendered quickly.

A system can pass every one of those checks and still fail the person trying to use it.

Accessibility testing during development remains essential, but production-oriented signals can help close the gap. Synthetic user journeys can tell us whether critical interactions still work. Design-system regression tests can protect shared components. Accessibility-aware release checks can help teams recognize when a deployment has affected a critical workflow.

For me, the principle is simple. If users cannot complete an important task because of an accessibility regression, I do not think we should describe the application as healthy just because the traditional dashboards are green.

That is not only an accessibility problem. It is a frontend reliability failure.

Build your own iPhone and Android app without coding for just $99.99

Cult of Mac

By: Cult of Mac Deals

9 September 2026 at 14:30

Photo of a hand holding an iPhone with a MacBook keyboard in the background, used to illustrate a post about Swiftspeed AI app builder

Get lifetime access to the Swiftspeed AI App Builder Pro Plan for $99.99 (regularly $1,440) and build native iOS, Android, and web apps without writing code.

(via Cult of Mac - Your source for the latest Apple news, rumors, analysis, reviews, how-tos and deals.)

The five important tools for controlling AI costs

InfoWorld

9 September 2026 at 10:00

We spent the last decade building entire finops departments just to decipher what the cloud bill was saying. Just when we figured out how to stop leaving idle compute instances running, generative AI introduced an infinitely more opaque, faster-moving layer of spend.

AI bill shock has spread like a slow moving hurricane across the industry. Aside from the spike in large language model (LLM) costs, the deeper architectural problem is attribution: where exactly is the money going, and exactly what value is it delivering to the business?

When API calls are wrapped in layers of automated agents and prompt templates, your application becomes a black box that consumes capital to spawn tokens. If a features engine is burning thousands of dollars a month just to have an LLM output pleasantries or process low-value data, that isn’t innovation, it’s an uncovered manhole, a gaping liability. Architectural maturity means treating tokens like any other constrained resource.

Fortunately, there are a few good levers we can pull to control AI spend. Here are the five key tools for taming the beast.

Model routing

Don’t use a nail gun if a thumbtack will suffice.

The most powerful models like the latest Claude Opus are some of the most sophisticated software systems ever built. They are capable of handling extremely complex and subtle use cases. They are also voracious beasts when it comes to compute (and therefore dollar) consumption. They are often overkill.

We naturally start with the most powerful model we have when prototyping and sketching out APIs and requirements. That’s because we are in the “get it working” mode and we don’t want to be tangling with model limitations. But for production, it is absolutely essential to dial in on what is the least powerful model we can use.

Sometimes we can address this by tuning the model choice and discovering what will work manually. But in larger enterprise architectures, we can move to a conditional routing layer. Frameworks like RouteLLM or Semantic Router can dynamically direct simple classification, basic text parsing, and user intent detection to fast and cheap utility models like GPT-4o mini or Claude 3 Haiku.

You can also use advanced AI gateways for routing and spend allocation. At the infrastructure level, wrapping this logic in a dedicated AI API gateway (such as Kong, Cloudflare AI Gateway, or Portkey) gives you the ultimate control. A mature gateway allows you to implement “cascade routing” to automatically fall back to a cheaper or open-source model if your primary LLM hits rate limits or latency spikes. More importantly, an AI API gateway centralizes your telemetry, stripping away the black box so you can enforce hard token budgets per tenant or per microservice. You are essentially building a load balancer for cognitive tasks.

Of course, AI gateways also introduce complexity and cost themselves, so these factors must be balanced into the equation. But the bottom line is, there is a right-sized model for your project. You can save a bundle by routing requests to a good-enough, Goldilocks model.

A side note: Another trend in model routing is “repatriation of compute.” That entails running models (often open-source models) on local hardware, instead of renting generative AI endpoints on the cloud. This implies all the trade-offs and benefits of other on-prem options.

Semantic caching

In conventional web apps, caching is a solved problem. You put Redis in front of your database, map an exact query to a cached payload, and you’re done. But LLMs break traditional caching because human language is infinitely variable. “How do I reset my password?” and “I forgot my login info” are entirely different strings, but communicate identical intents.

If you rely on overly precise matching, your hit rate will be effectively zero, and you will pay the full token price for every phrasing permutation.

Enter semantic caching. It’s like regex for meaning. Instead of hashing the raw text string, you run the incoming prompt through a fast embedding model and do a similarity search against previously answered prompts. You define a confidence threshold (say, 0.92), and if it is met, you serve the cached response (bypassing the LLM and its cost entirely).

The architectural effect is twofold. First, a semantic cache hit reduces your inference cost for that query to exactly zero. Second, it drops response latency from seconds to milliseconds. You can implement this using purpose-built frameworks like GPTCache, or, if you prefer a leaner infrastructure, by using pgvector inside your existing PostgreSQL database to store and match the embeddings.

However, just like dynamic routing, semantic caching demands a rigorous architectural calculus. You are trading generation costs for embeddings and lookup costs. To check the cache, you still have to tokenize the prompt, call a cheap model (like text-embedding-3-small), and execute a vector search.

You also introduce the very real danger of semantic flattening. Tuning the similarity threshold is a delicate art. Set it too low, and your application starts serving general, recycled answers to nuanced user questions. For open-ended, creative, generative AI applications, semantic caching is practically useless. But for retrieval-augmented generation (RAG) implementations, customer support bots, and internal knowledge bases where users ask the same 20 questions a thousand different ways, it is the most effective cost-reduction lever you can pull.

Prompt caching

Whereas semantic caching stores the response to a given intent, prompt caching stores the data needed to contextualize the question. When the user enters a prompt, it is sent to the generative AI endpoint. But instead of your application having to repeatedly gather and send the massive contextual info needed to frame the prompt—from a RAG pipeline, a database, or other source—that information is already pre-loaded in the context cache. The model simply applies the new question to the cached data, slashing both your latency and your input costs.

The core idea behind prompt caching is that the AI engine itself retains the relevant contextual data for incoming prompts. Tokens cached this way get a massive discount (50% to 90%) compared to raw input tokens. But the details of how it is handled vary between providers.

For OpenAI, prompt caching is automatic, but requires attention to how your app uses the endpoints in order to leverage them. OpenAI enforces a 1,024-token minimum threshold. More importantly, it requires an exact, byte-for-byte prefix match starting from token zero. If you inject a dynamic variable (like a timestamp or a user ID) at the very top of your system instructions, you instantly invalidate the cache and pay full price for every token that follows.

Some other providers, like Anthropic Claude and Google Gemini, use explicit caching. You must deliberately engineer the cache into your architecture, either by hitting a dedicated API endpoint to upload the static document (the context data) with a defined time to live, or by injecting strict cache control breakpoints into your JSON payload. Explicit caching often requires paying a “write premium” up front, but it gives you the control to guarantee your heavy RAG data stays alive in memory for hours rather than minutes.

Prompt discipline

Just because a model can consume gargantuan amounts of tokens, doesn’t mean it should.

With models now capable of ingesting one million to two million tokens in a single prompt, there is a temptation to brute-force use cases into submission by dumping the entire PDF, the entire codebase, or the last three years of logs into the payload and let the LLM sort it out.

This is a finops blunder. You are paying for every single token you send. Furthermore, from an engineering standpoint, stuffing the context window runs into the very real “muddy middle” problem (aka “context rot”), where the model loses resolution into the middle of the content. As the payload grows, performance and accuracy suffer.

The cure is a strict RAG diet. Naive RAG implementations might pull the top 20 results from a vector database and blindly concatenate them into the prompt. Architectural maturity requires a more ruthless filtering process before those tokens ever reach the expensive LLM.

This is where reranking becomes key. Reranking takes the compiled prompt context and makes sure it is efficiently structured before it hits the LLM. Instead of sending raw search results directly to your primary model, you introduce a tiny, highly-efficient cross-encoder (like Cohere Rerank or an open-source BGE model). You cast a wide net in your cheap vector database to fetch 50 potential matches, then pass them to the reranker to score their actual relevance against the user’s specific prompt. The reranker ruthlessly trims the fat, allowing you to pass only the top three or four hyper-relevant chunks to the expensive frontier model.

You are trading a massive, expensive LLM context payload for a tiny, cheap reranking step. The architectural math here is almost always a resounding win. While a cross-encoder might add 100 milliseconds of latency to the pipeline, it routinely slashes prompt token counts by 80% or more, while simultaneously increasing the accuracy of the final generation. In the generative AI spend playbook, giving your LLM less to read is one of the highest-leverage moves you can make.

Response constraints

By default, LLMs are polite, conversational, and verbose. If you ask an API for data, it desperately wants to start with “Certainly! Here is the JSON you requested,” and end with “Please let me know if you need anything else!”

In a web interface, this is charming. In an application’s API architecture, it is a leak in the finops hull. Generation tokens (output) are almost universally priced 3x to 5x higher than prompt tokens (input). Every pleasantry is burning your most expensive commodity.

Taming LLM responses requires strict API-level enforcement. First, max_tokens must be treated as a hard circuit breaker to prevent runaway generation loops, not as a formatting tool. (Relying on max_tokens to shorten answers often just results in truncated, unparseable JSON.) Second, use stop sequences (stop_words). If your model only needs to output a SQL query, setting a stop sequence for the markdown closing tag (“`) or the word “Explanation:” forces the model to sever the connection the moment the computational work is done.

Modern APIs also offer structured outputs (or strict JSON mode), which forces the model to adhere to a rigid schema you define. By combining this with zero-tolerance system prompts (“Output ONLY valid JSON. Omit all conversational text and pleasantries. Failure to do so will break the system”) you force the LLM to behave like a traditional API endpoint rather than a chatbot.

The architectural balance here is to avoid degrading the model’s reasoning ability by strictly limiting its input. LLMs do not have internal working memory, they “think” by iterating on token output. If you force a model to solve a complex logic problem and output a single boolean true or false without giving it space to reason first, its accuracy will plummet. The finops goal isn’t really to force minimal words; it is to ensure that every expensive output token is performing actual computational work (like chain of thought reasoning).

For more on how to manage AI back ends, take a look at this article.

How to effectively consume generative AI to satisfy user needs is the first battle. The second is keeping the finances from undermining that hard work. The five keys discussed here should give you all the power you need to make sure the engineering and the cost controls are working together.

Suddenly I’m a Go developer

InfoWorld

9 September 2026 at 10:00

I guess I’m a Go developer now.

My first programming language was BASIC. My seventh-grade math teacher, Mr. Schulte, taught BASIC to us on Fridays. He gave us programming assignments for homework during the week. Some of us loved it so much we stayed after school to play games and create extra programs, including our own crude, text-based adventure games. I remember hearing a story about a crazy new language called Pascal that didn’t have line numbers. How in the world would that work?

Years later I became a Pascal programmer — an ardent one. So ardent, in fact, that I would argue for hours with Visual Basic developers on CompuServe, AOL, and Usenet, telling them how Turbo Pascal, and then Delphi, was clearly a superior language. I was completely self-taught — if you count all the help that others gave me as teaching myself. I wrote shareware, became pretty competent, and eventually started helping others.

People would tell me I should branch out and learn other languages — C++ or C#. But I wanted to be a master of one instead of a jack of all trades. Besides, we were all Windows developers and there was no better tool for writing Windows applications than Delphi. It became a point of pride for me, and I would stick up for Delphi and its company Borland no matter what. It got so bad that my friends would tease me, calling me “Baghdad Nick” after the famously optimistic Iraqi foreign minister during the Iraq War.

Delphi by day

Looking back, I was foolish. I did become a pretty good Object Pascal programmer, but I would have benefited from learning JavaScript or a functional language. I did eventually learn TypeScript and a bit of C#. (I guess I was following Anders Hejlsberg around…)

But if you were to ask me to this day what kind of code I write, I’d say I’m a Delphi developer. Indeed, that is my day job. I understand Delphi deeply, knowing all the nuances and insider tricks to making the language my own.

But over the course of a day or two, I’m now a Go developer, too.

If coding all day isn’t enough, I have taken on a side project for a friend. I’ve been writing Delphi code for him for a while, but now he wants to shift gears. He has asked me to build a cross-platform application for his company written in Go and Wails Version 3. Now, I’m no Go expert. In fact, I only have a passing familiarity with the language. I can decipher what is going on, but I’m far from understanding the intricacies of what makes Go tick.

But that doesn’t really matter anymore, does it?

Claude bootcamp

I asked Claude to give me a tutorial on Go and Wails. It did a great job, quickly bringing me up to speed on the basics. (How cool is it that you can so easily turn a struct into a JSON object?) I described what I wanted, and asked for recommendations for libraries. The project uses Go for the “back end” and React on the “front end.” (I use the quotes because those “ends” are actually all built into a single binary.)

In about an hour I came to understand the architecture and got a basic application up and running. I told Claude what I wanted, and it obliged. I described the basic data structures and objects needed, and Claude wrote them quite elegantly. I didn’t actually write a line of code, but there the project was, in all its glory. Go, TypeScript, and HTML.

Claude managed the Git repository and even created a nice little CI/CD pipeline for me, building binaries for Mac, Linux, and Windows, all in one step.

I went from being basically ignorant of the Go language to having a moderately sophisticated application in a day.

The thing is, I’m still not that good at Go. Not that good at all. But I’m a Go programmer nonetheless. Maybe next weekend I’ll become a Python programmer.

AI won’t kill SaaS

InfoWorld

8 September 2026 at 10:00

Nvidia CEO Jensen Huang argues it’s the “most illogical thing in the world” to believe AI will kill SaaS, and says the markets “got it wrong” in sending SaaS stocks tumbling. Why “illogical”? Because, as analyst Benedict Evans stresses, the fact that everyone can now easily spin up code actually doesn’t solve any SaaS problems, because creating code and tools is the easy part: “The hard part is knowing that you need a tool for this in the first place and then knowing what the tool should do.”

If that’s true, then what Shopify is doing with Sidekick is potentially revolutionary by embracing and extending the traditional SaaS model.

In December, Shopify introduced custom app generation through Sidekick. A merchant (Shopify customer) describes a tool, and Sidekick writes the code, using Shopify’s interface components and connecting to its Admin API. It’s a cool option and has been very popular. Shopify says merchants created almost 4,000 custom apps in the first three weeks following the release.

This isn’t surprising: Shopify is delivering the customization customers have long wanted from enterprise SaaS while ensuring they still enjoy the comfort of ongoing support and platform stability.

I suspect we’ll see more enterprise SaaS like this, despite silly predictions that AI would kill SaaS. Cheaper code gives customers a way to fix the things they dislike about an application without replacing everything they do like. For a vendor willing to accommodate that, AI could make its product considerably more useful and, hence, much “stickier.”

The feature that never ships

Cloudflare’s Jeremy Morrell discusses this in an excellent essay on extensible software. “In the past year, your users have suddenly acquired the ability to speak code into existence,” he writes, allowing for “markets of one.” His argument is that software should let customers put that ability to work, building additions around an established core.

Morrell has a commercial interest here, which he discloses: He works at Cloudflare and thinks its infrastructure is well suited to running those additions. He’s not wrong. But it’s the general opportunity to match customizability with stability that matters so much.

There are really (really) good reasons why vendors haven’t allowed customizability of their core products. Enterprise software companies limit deep customization to prevent high maintenance costs, security risks, and broken system updates. Oh, and to prevent their products from becoming convoluted bloatware. No matter how much you may want feature X, that feature would likely be irrelevant to almost everyone else, so adding it would complicate the product for the majority. No one wants that.

Customers have long paid consultants to customize software or they’ve hired developers to do it themselves (all while taking care not to violate their vendor licenses). Morrell points to Salesforce, which has spent years letting businesses build their own logic into its platform. In other words, AI didn’t invent extensibility. No, AI just makes it incredibly cheap. There’s a meaningful opportunity here for customers who can describe the change they need but have never been able to get it built. Some will save money on software or customizations they previously bought, while others will finally get software that nobody could profitably sell them.

Following the money

This change, however welcome for enterprises, won’t be comfortable for developers selling a narrow reporting tool or a small workflow improvement. Their customers may decide that an LLM-generated extension is good enough. An AI-built customization needn’t match every feature of the commercial product. It only has to do the job that the customer was paying for.

The platform underneath that extension has a different calculation. Every useful tool built against its data and APIs gives the customer another reason to stay. AI can make an individual feature less valuable to sell while making the product that hosts it more valuable to own. (Yet another reason to tsk-tsk foolish eulogies for SaaS companies.)

Ultimately, this should benefit anyone who both knows the business and can build more efficiently for that business. Benedict Evans makes this point in his newsletter, arguing that cheaper coding doesn’t solve the problem of discovering what software ought to do or persuading an organization to use it. A useful idea that spans several departments still needs those departments to agree. Even excellent code can’t sign off on a new way of running accounts payable.

Evans also describes how improvised tools eventually become important enough to need formal ownership and support. Anyone betting that companies will happily maintain everything they can generate should spend some time with the person who inherited the last departmental spreadsheet. Silicon Valley folks with little knowledge of real-world enterprises might think enterprises will vibe code all their software now, but they won’t. That’s not how software works because it’s not how people work.

This cuts both ways for established software vendors. Customers may continue to need their expertise while becoming less willing to accept a product’s limitations. A vendor that makes customization easy can benefit from that impatience, while one that blocks it has given the customer a new reason to investigate alternatives. The difficult decisions will come when customization interferes with what the vendor already sells.

I wrote in 2022 about Microsoft’s complicated relationship with open source, and how a company’s broad commitment to openness can collide with a division’s revenue targets. The same tension applies here. A CEO can be excited about letting customers build anything. The executive responsible for selling the premium reporting package may be less enthusiastic. Resolving that disagreement will determine what customers actually get to build, however capable the AI becomes.

Getting comfortable

Customers have their own calculations to make. An application shaped around the way your company works can be wonderful, but it can also be a pain to leave.

There are limits to what Shopify’s generated apps can do today. They are admin tools, not storefront or checkout extensions. Shopify also tells merchants to test them before installation because they can change customer-facing data. A generated app can be wrong, and running it inside a familiar product doesn’t relieve the merchant of checking its work.

Shopify lets merchants inspect and edit the generated code and restore an earlier version. That’s useful, but code written for Shopify’s Admin API still depends on Shopify. Having the source doesn’t give you another company ready to run it unchanged.

AI might eventually make moving those extensions easier, too. But recreating a screen is one thing; reproducing the behavior it relies on throughout a business is a considerably bigger undertaking. That includes the assumptions about records and transactions that the custom code gets to take for granted while it runs on the original platform.

None of this makes customization a bad deal. Customers can gain more control over their daily work while, somewhat ironically, becoming more dependent on a supplier. For SaaS vendors, that’s a good commercial reason to let customers finish the bits their developers will never get around to building, even when it costs them an add-on sale (and potentially some support headaches, depending on how they’re architected for extensibility). The customer is doing work that makes the underlying product more valuable.

AI, in short, has the potential to make SaaS dramatically more valuable, not extinct.

Ternus and the App Store

Michael Tsai

By: Michael Tsai

7 September 2026 at 19:17

Mark Gurman (tweet, John Gruber, Mastodon, 9To5Mac):

The Schiller exit is a bit more notable. Unlike Maestri, he still had a real job: The veteran executive ran the App Store and Apple Events. The App Store portion alone is no joke. It generates an estimated $30 billion a year and can be very challenging to manage. Apple’s app marketplace is loved by many consumers, but often criticized by developers and subjected to increasingly onerous regulations. It was a busy, all-encompassing gig and nowhere near a semiretirement.

[…]

But there’s a bit more to the story, I’m told. Ternus and services chief Eddy Cue want to make even more money from the App Store and figure out ways to raise margins and squeeze additional recurring revenue from the platform. Schiller, on the other hand, seems to believe that such moves will only further irk developers and governments. While there was no internal blowup or anything like that, it’s something he wanted no part of.

“Product guy” dreams about margins, forgets that people buy Apple hardware because of the software.

Tim Hardwick:

Apple admitted during its July earnings call that regulatory changes have started to weigh on its Services growth. Apple reported services revenue of $30.7 billion for the June quarter, a record for the period but short of the $31.4 billion that analysts expected.

Patrick Smith:

The hardware guys treating third-party developers like they do their suppliers has stifled Apple’s software platforms

M.G. Siegler:

I thought that perhaps with the move to John Ternus as CEO, it could also be an obvious time to revisit such policies. But I hadn’t considered that Apple could actually be in favor of making them more onerous.

See also: Mario Guzmán and Indie Simulator.

Previously:

What it took to triple our software engineering output in 18 months

InfoWorld

7 September 2026 at 10:00

A year and a half ago I sat in a conference in Silicon Valley and watched an AI agent write production-quality code in real time. I knew our engineering model was going to change, and within a few weeks my team was past the question of whether to adopt AI. The real debate was how far we were willing to go in redesigning how we build legal software around it.

Since then, output per engineer across our R&D organization has roughly tripled. We ship nearly twice as many releases per quarter, deployments have grown from 82 to more than 155, and customer reported defects are down 65% per million lines of code across the last 18 months. We track all of this with DORA metrics, cycle time, pull requests merged per developer, and lines changed per developer against a fixed baseline, and the improvement holds up on every measure.

Getting there had less to do with the tools than with how the organization works. Every CTO I talk to is running some kind of AI coding pilot and seeing decent results: faster prototyping, better test coverage, quicker first drafts. Almost none of them have changed the development life cycle itself, and that is where the real gains were for us.

We spent the past 18 months rebuilding our product development life cycle around AI. Requirements, development, testing, security, deployment, and governance all changed, some beyond recognition. We got plenty wrong along the way, and a few of the lessons would have saved us months if someone had handed them to me at the start.

The biggest gains came from removing handoffs

We assumed most of the productivity gain would come from AI writing code faster. It didn’t. The biggest gains came from getting rid of the handoffs between stages.

In the old model, a feature passed through at least four handoffs on its way out the door: product to development, development to QA, QA to security, and security to deployment operations. Each of those queues cost us time and context. A feature could be code complete in a day and then sit for two weeks waiting on teams that each had their own backlog and their own priorities.

So, we restructured. A team now carries a feature end to end, and AI agents handle the rote work at each stage. A feature that used to take 15 days to go from spec to a working pull request now takes about four hours.

If you take one thing from this article, measure the total time from idea to deployment rather than lines of code generated, and go find where the work sits idle. That is where the gains are hiding.

Governance accelerated adoption

Most engineering organizations add governance after the innovation has proven itself. We flipped that order. We build software for lawyers, where a wrong answer on high-stakes work can cost a client dearly, so moving fast and cleaning up later was never on the table.

Before we scaled past the first pilot teams, we set standards for code review, security scanning, and quality on anything AI produced. Confidence scoring lets routine approvals pass through automatically while anything below threshold goes to a human. Security and quality gates run inside the pipeline itself rather than as a checkpoint at the end.

What we didn’t expect was how much this sped up adoption. Engineers trusted the system, so nobody had to wonder whether code from an agent would pass review, introduce a vulnerability, or get them blamed when the model got something wrong. Eighteen months ago, about 3% of our pull requests had any AI assistance. Today it’s 68%, every engineer in the organization uses AI, and vulnerability density has fallen 76% over the same period. Governance came first and adoption followed, which is the opposite of the sequence most people assume.

Pick your tools and get good at them

A better model or agent ships almost every week, and early on we burned real energy evaluating them. At some point we stopped and standardized on a small set of coding agents.

Going deep paid off more than staying current. We learned the strengths and failure modes of our coding agents and tuned our workflows around how they actually behave, which matters more than whatever the next release promises. Teams that chase every new model spend all their time evaluating, and every switch resets the muscle memory and the trust they built with the last tool. They end up with a lot of pilots and not much change.

Automate the entire life cycle

The obvious place to point AI coding tools is at the coding. Some of our biggest wins came from the edges of the life cycle.

We built or deployed dedicated agents for every stage: requirements, sprint planning, code generation, test authorship, security scanning, deployment, and site reliability. Each one owns a single job and gets measured on a single outcome. What makes it work is a centralized knowledge repository that product and engineering maintain together, a curated and structured knowledge base with retrieval-augmented generation on top. Every agent pulls product context, domain rules, and prior decisions from that same source of truth, so what comes out of one stage is clean input for the next. Without it we would have a set of fast agents that lose the thread at every handoff.

Requirements definition was the most dramatic early win. It used to take weeks. We built an agent that grills a raw product idea the way a tough stakeholder would, pushing on gaps until the spec is ready for development, and that phase now finishes in an afternoon.

Testing moved even faster as quality engineers shifted their focus to optimizing and improving the agents that build tests. AI now generates 99% of our new tests, up from 47% a year ago, with more than 39,000 AI-developed tests in the suite, a number we could never have staffed our way to. Each test is generated directly against the code change it covers and checked against the functional requirements defined in the product requirements document, so growth in volume doesn’t come at the cost of traceability. Engineers spend their time reviewing, refining, and hardening what the system produces rather than writing first drafts, and that shift, from authoring tests to curating and strengthening them, is a big part of why quality improved while we sped up.

We started with one area of the product development life cycle, got it right, then expanded. I would recommend that sequence to anyone. The key is to develop a model that can be expanded to the rest of the phases after experimenting on what works rather than trying to roll it out to every area and having to rebuild the framework later.

Design for the velocity you’re about to create

We set a public and very specific goal: double R&D productivity in 12 months. It was uncomfortable, and there were stretches where I wasn’t sure we would make it. But a number that aggressive forced everyone to accept that incremental improvement wasn’t going to get us there, and we had to take apart old ways of working while the technology underneath was still maturing. We hit the goal inside the year, and output has kept climbing ever since.

What I underestimated was how fast the bottleneck would move downstream. Once engineering sped up, the constraint moved to go to market. Every release still needs documentation, enablement content, briefings for customer success, and customers who are ready for what’s coming, and for a while we were shipping faster than any of that could keep up. We now treat go to market as its own automated phase of the life cycle, with the same agent-driven approach we applied to code and testing, and a simple rule that nothing ships until the rest of the organization can support it. If I were starting over, I would build that capacity from day one instead of finding out about it the hard way.

The real transformation is organizational

The tools matter, but everyone can buy the same tools. What we actually did over these 18 months was redesign the system around AI: the way work flows between teams, the governance that lets people trust the output, the small set of tools we committed to, and a target aggressive enough to force all of it. Any one of those alone would have produced another interesting pilot. Together they produced a different engineering organization.

None of this is finished. The models keep improving and the life cycle will keep changing with them. The organizational groundwork is what lets us absorb each improvement as it arrives, and for any technology leader still stuck at the pilot stage, that groundwork is where I would start.

—

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.

Why AI-first development matters — and how to get there

InfoWorld

7 September 2026 at 10:00

For a growing number of organizations, AI has become not only a key component of the software development process, but the centerpiece.

These enterprises have shifted toward an AI-first development strategy, where artificial intelligence is integrated into all phases of the software development life cycle, rather than being a feature to use selectively during the process.

In most cases, this means a move toward agentic workflows, with developers acting as architects responsible for overseeing AI agents. But in a larger sense, AI-first development is a transformation of how developers explore and understand code, with a focus on designing and producing more and more applications with AI in mind.

Why AI-first development?

For what is surely a rising number of software development teams and organizations, the AI-first approach is becoming more necessary than optional.

“AI-first development matters because the alternative is increasingly untenable,” says Mona Rajhans, senior software engineering manager at cybersecurity products provider Palo Alto Networks, who leads a 14-person AI engineering team. Simply put, development teams that use AI to build applications can accomplish far more in less time.

By the same token, it is increasingly important to design applications with AI components and agentic users in mind. “Bolting AI onto an existing architecture after the fact is like adding a second floor to a house with no load-bearing walls,” Rajhans says. “It works until it doesn’t, and when it breaks it breaks badly.”

One of the biggest advantages of AI-first development is the speed with which teams can complete projects and get products into development. This includes identifying and addressing issues much earlier than would be otherwise possible.

“Work that used to stall in the backlog for quarters, like platform migrations and updating aging libraries, now clears in days,” says Collin Hogue-Spears, senior director, DTE, at application security software company Black Duck Software. “Defects surface at design time rather than in production, where a catch costs a code review instead of an incident.”

When you apply AI from the onset of your software project, “you’ll typically learn about issues with your concept in hours rather than weeks,” says Bob Hutchins, CEO at consulting and advisory firm Human Voice Media. “I developed full production applications where the first working prototype was completed prior to the kickoff meeting for a traditional shop.”

AI-first development also helps to keep unnecessary costs down. “Retro-fitting AI into projects is often costly and difficult,” Hutchins says. “Project designs that include AI from day one result in cleaner data flow patterns, much clearer permission schemes, and user interface elements that understand how agents will interact with these systems.”

Implementing a chatbot to an existing application produces a chatbot, Hutchins says. “Designing for AI provides an entirely new type of product,” he says.

Perhaps most important, using AI-first development causes teams to improve their documentation and overall clarity in their thought processes, Hutchins says. “Since an AI cannot read minds, teams which follow this approach will provide better documentation specifications, because the specification itself will serve as the actual instruction set.”

How to succeed at AI-first development

Not all development teams are ready for this sea change in the development process. Many struggle to adopt AI-first development for any number of reasons. Maybe they have underlying engineering systems that rely on undocumented knowledge, implicit rules, or manual oversight. Or they don’t have the right quality assurance, security, and testing processes and skills in place.

AI requires precise context in order to function correctly, and teams that rely on undocumented legacy codebases and obscure architectures might find that AI generates inaccurate information that ends up requiring significant rework by humans.

By following a few good practices, developers and teams can prepare for an AI-first environment and excel at this new paradigm for development.

Add new skills and roles

One of the most important steps is to add new skill sets and roles that fit into the AI-first strategy, even if that means uncomfortable changes for some. Without the needed skills in place, nothing else in the AI-first approach will work well.

“The scarce ability is no longer writing code, but reading it quickly and assessing it accurately,” Hogue-Spears says. “The strongest developers will work as architects: they frame the problem, direct the agent, and verify the result. That is how one engineer safely oversees the work of several.”

Organizations need to focus staffing on the senior architecture roles and cut or redesign the entry-level junior coding roles, says William Flaiz, founder of CleanSmartLabs, a provider of data cleansing products.

“Tools like Claude Code can write good clean code fast so that is no longer the bottleneck,” Flaiz says. “The judgment and experience of what to build, how to architect it, and what is going to break comes from the judgment of senior architects. Skip this and you will ship product fast, but it will break under user load.”

Consider training programs to ease the transition

Much of AI-first is relatively new for a lot of organizations, and many developers and managers will need to learn what it means to be architects in this new world.

The architect role “is changing in a way most organizations have not caught up to yet,” Rajhans says. “The job is less about writing code and more about designing the boundaries, where the agent acts, where the human decides, and what happens when the agent is wrong. That last question is the one most teams skip until it is too late.”

Development teams also need to shift the developer role to an orchestrator role for agentic development, Flaiz says. Training will be needed in a lot of cases.

“They have enough product context and user understanding to where agents hand off, conflict, and which agent wins when in conflict,” Flaiz says. “This is not a programmer role in the traditional sense, but a new role that is a good transition for junior and entry-level developers.”

In addition, Flaiz recommends integrating user experience expertise into the development process, “not as a starting point only or as a gate check at the end,” he says. “AI can write requirements, specs, and assist with architecture, but it doesn’t understand your users. It doesn’t know the habits of how real users interact with software. That comes from watching [user experience] tests, analyzing behavior, and being human.”

Shift toward agentic workflows

A key part of AI-first development is shifting toward agentic workflows, where autonomous agents deploy reasoning, planning, and external tools to achieve complex goals. This is the scenario in which developers will act as architects who oversee AI agents as they work to solve problems and build applications.

“The fundamental transition here is from typing—writing code—to specifying what you want done,” Hutchins says. “People who thrive within agentic workflows are individuals who can clearly define systems, think through potential [uncommon scenarios], and assess others’ work objectively.”

While keying in code might seem to matter at times, the skills required for agentic workflows “represent true architecture and editorial judgment,” Hutchins says. “As such, senior engineers’ values increase rather than decrease.”

With the new workflows, “developers will spend less time writing every line of code manually and more time defining goals, considering product constraints, iterating through architecture, designing intuitive user interfaces, envisioning data flows, and calculating expected outcomes,” says Bob Brauer, founder and CEO of Interzoid, a consultancy and provider of data enrichment and structured intelligence services.

“AI tools and agents can then help generate code, test plans, documentation, deployment scripts, and monitor ongoing maintenance requirements,” Brauer says.

Create new testing and review processes

Another good practice is to create a new review process prior to building agent pipelines. “AI will generate ‘plausible’ garbage without hesitation,” Hutchins says. “Organizations that are currently suffering damage due to [poor] generation are the same ones that grew their generation capabilities faster than their ability to verify generated output. We have already seen consultants return client monies paid for fake citations. This represents a process failure, not a technology failure.”

Reviews need to be ongoing. “Design for the human checkpoint,” Hutchins says. “For every workflow the questions should be ‘where does a person with judgment need to review before we send this?’ If you cannot answer this question for your current workflow, then you are not implementing AI-first development. You are simply allowing unsupervised delegation.”

As for testing, “developers have to get comfortable with non-determinism,” Rajhans says. “AI-first systems do not behave the same way twice. Testing, observability, and failure handling all have to be redesigned from scratch around that reality. Teams that treat AI outputs like function returns get burned fast.”

Start small and proceed with caution

Despite the urgency among many business leaders to move aggressively toward AI, a somewhat cautious approach might be best for both near-term and long-term goals.

“Begin small and track everything,” Hutchins says. “Allow developers access to approved tools and clearly defined guidelines, then determine where [and] when AI reduces developer time verses when AI generates additional rework.”

AI-first development will become increasingly critical as competition intensifies and software teams face growing pressure to deliver faster, more sophisticated, and easier-to-use applications, Brauer says.

“By leveraging AI across requirements gathering, architecture/design, coding, testing, deployment, and ongoing maintenance, development teams can significantly increase their effectiveness and create greater organizational value,” Brauer says. “Organizations that embrace AI-first development will be better positioned to deliver higher-quality software faster and gain a meaningful competitive advantage.”

Machine Vision Joins Apple’s U.S. Factory Training Push

AppleMagazine

By: Ivan Castilho

5 September 2026 at 21:32

Machine vision is becoming a practical part of Apple’s U.S. factory training strategy, giving smaller manufacturers access to techniques that help production systems inspect what they make. The recent opening of the Houston manufacturing training center extends that effort into a working electronics production campus, while courses developed with Michigan State University introduce businesses to image analysis, machine learning, and automated quality checks.

For a manufacturer, the attraction can be straightforward: finding a missing component before a product leaves the line, identifying a surface defect before more material is added, or recognizing an assembly error while it can still be corrected cheaply.

These are manageable problems with measurable costs. They also offer an accessible starting point for businesses interested in AI but unsure where it belongs inside their operations.

Machine Vision Enters the Manufacturing Curriculum

The Apple Manufacturing Academy’s current course catalog includes Machine Learning with Vision, a course introducing machine learning and deep learning through practical activities using Apple devices. It sits alongside instruction in automation, quality fundamentals, manufacturing data, maintenance, and process improvement.

That placement gives the subject useful context. A camera can produce an image, but a factory needs a decision: accept the part, flag it for review, check its position, or investigate a process that may be drifting out of tolerance.

Machine vision connects image capture to those operational decisions. Some applications use established rules to measure dimensions or identify patterns. Others use machine learning to recognize acceptable variations and defects that are difficult to describe through fixed rules.

The academy’s public description confirms introductory activities with Apple devices. It does not identify a particular iPhone, Mac configuration, software framework, or industrial camera package that every participant must use.

For smaller businesses, the educational value is learning how to define an inspection problem before purchasing equipment. A company that understands what it needs to detect can have a much more productive conversation with an automation supplier.

A woman in a pink smartwatch uses Visual Look Up on her smartphone to snap a photo of her food at a wooden table, with plates of rice, salad, and grilled meat in front of her.

Houston Brings Training Closer to Production

Apple opened its 20,000-square-foot Advanced Manufacturing Center on August 13. According to the Houston opening announcement, the free program offers small and medium-sized businesses instruction in advanced manufacturing, interactive labs, and experience with representative production equipment.

The first cohort worked with machine learning for quality control and advanced automation. Participants also explored production challenges, used a holographic table, and assembled and laser etched a product.

The location adds context: the center occupies the same Houston facility that builds advanced AI servers and is scheduled to begin Mac mini manufacturing this year. Its teaching program includes final assembly principles and design considerations for printed circuit board assembly.

Houston is the company’s second U.S. manufacturing learning site, following the Detroit academy established with Michigan State University in August 2025. By the Houston opening, Apple said the academy had helped nearly 1,000 workers, students, and entrepreneurs.

That figure measures people reached through the program. It should not be interpreted as 1,000 factories already operating AI inspection systems.

A diagram titled "Adapters" displays various blue bars labeled Summarization, Proofreading, Mail replies, Tone adjustment, Role Playing, Query handling, Friendly, and Urgency. Below these bars is an icon representing a neural network structure powered by Apple Intelligence Servers. — Apple Machine Learning Adapters

A Useful Inspection Starts With a Specific Problem

Imagine a small business assembling control panels. Each panel needs the same set of connectors, positioned correctly before the enclosure closes. A machine vision station could check for a missing connector while the assembly remains accessible.

The value comes from catching that error at the right moment. Discovering it after packaging creates extra handling and rework. Discovering it after delivery may involve a return, a service visit, and an unhappy customer.

A useful pilot would therefore begin with that single inspection point. The business could compare the existing process with the assisted one, recording missed defects, unnecessary rechecks, inspection time, and the cost of correcting mistakes.

The camera is only part of the exercise. Workers must agree on what constitutes a defect, collect representative examples, and decide what happens when the result is uncertain.

Those decisions draw heavily on experience already present on the factory floor. An operator may know that an unusual mark is harmless, while a barely visible change in alignment predicts a later assembly problem. Turning that knowledge into consistent inspection criteria is valuable work in its own right.

These are examples of how manufacturers could apply the training, rather than projects Apple has announced for individual participants.

Teaching Machine Vision to Handle Ordinary Variation

A demonstration can look impressive when every part arrives clean, centered, and evenly lit. Production introduces variation: different surface finishes, reflections, changing positions, and batches that look slightly different while remaining within specification.

The inspection system must distinguish those acceptable differences from actual failures. Otherwise, it can reject good products or allow defective ones to continue.

Industrial vision supplier Cognex’s deployment guidance recommends validating systems across product models, defect types, and manufacturing lines. It also describes an approach in which automated inspection handles preliminary checks while experienced inspectors review borderline cases.

For a business learning machine vision, that offers a realistic way to introduce the technology. Employees can compare its decisions with their own, investigate disagreements, and determine whether the results justify extending its role.

An inspection system also needs an owner after installation. Someone must understand when a product change requires new examples, when equipment needs attention, and when performance has slipped. Training that responsibility into the workforce helps prevent an initially successful project from becoming an unreliable fixture.

Smaller Manufacturers Are Already Applying the Training

The academy’s reach extends beyond companies producing consumer electronics. In its May update on participating manufacturers, Apple highlighted Block Imaging, a Michigan business that services and refurbishes medical imaging equipment.

The company hosted visitors during the academy’s Spring Forum and described applying techniques learned from Apple engineers and Michigan State experts to its operations. Apple reported that the academy had supported more than 150 American businesses at that point.

The announcement did not identify a particular machine vision deployment at Block Imaging or publish a percentage improvement attributable to visual inspection. It does provide an example of the training reaching businesses whose work differs substantially from building an iPhone.

That wider participation could benefit Apple indirectly. More capable domestic manufacturers can support equipment maintenance, tooling, components, and specialized production services. Participation itself does not establish a supplier relationship, but the skills can be used across customers and industries.

Access Without a Large Training Budget

The academy’s participation information says eligible small and medium-sized U.S. manufacturers can take courses free of charge. Online learning includes on-demand materials and interactive modules, while in-person sessions add workshops and more personalized instruction.

Digital badges are available for different topics, and completed course materials remain accessible for review. That gives a business room to train an employee around existing production commitments and return to the material when an actual project begins.

Free instruction does not remove the eventual cost of cameras, lighting, integration, maintenance, or employee time. It can help a manufacturer understand those costs before committing to a system.

The first proposal brought back to the factory might be modest: inspect one recurring defect at one station and record the results for several weeks. A supervisor can then weigh the time saved against the extra reviews, maintenance, and mistakes. That is the kind of evidence a business owner can use when deciding whether to install a second station.

A lineup of Apple devices—a Vision Pro headset, iPad, MacBook, iPhone, and Apple Watch—displays colorful swirling dot patterns on their screens against a white background, highlighting seamless Apple Account recovery across the entire ecosystem. — Image Credit: Apple Inc.

The post Machine Vision Joins Apple’s U.S. Factory Training Push appeared first on AppleMagazine.

Sam Altman Calls Himself a ‘Mega Apple Fanboy’ Amid Lawsuit

AppleMagazine

By: Jack

5 September 2026 at 14:35

Sam Altman has described himself as a “mega Apple fanboy” while discussing the trade secrets lawsuit now dividing two companies that were close technology partners only two years ago.

Speaking with journalist Alex Heath on the Sources podcast, the OpenAI CEO said he was “very sad” when he learned about the case. His admiration for Apple, however, did not soften his rejection of the allegations against OpenAI and two former employees.

Altman said the company does not want another business’s intellectual property and would dismiss anyone found to have improperly brought confidential material into OpenAI. He also said an internal investigation left him convinced that the employee at the center of the dispute had not acted improperly.

That is OpenAI’s account, not a resolution of the lawsuit. Apple alleges that trade secrets connected to hardware design, manufacturing and supply-chain operations were taken for the benefit of OpenAI’s consumer-device program. The defendants deny wrongdoing, and the court has not settled those competing claims.

Sam Altman Separates Personal Admiration From the Case

Altman’s response was unusually personal for a dispute of this scale.

Technology executives normally leave active litigation to lawyers and carefully written company statements. Calling himself a fanboy places Altman on familiar ground with millions of customers who admire Apple products while disagreeing with a corporate decision.

His first reaction, he said, was to assume that the accusations sounded serious enough that someone must have behaved badly. OpenAI then examined the matter and reached a different view.

Altman said the company would act if it found that someone had taken another company’s intellectual property. At the same time, he argued that an employee who had done nothing wrong deserved a defense.

That distinction is central to OpenAI’s public position. It does not deny that former Apple employees now work on its hardware initiative. Instead, the company disputes the allegation that their movement between employers brought protected information into the program.

Personal admiration cannot answer the evidentiary questions. Being a customer, praising a company or enjoying its products has no bearing on whether confidential files were accessed or used. Altman’s comments are more revealing as an account of how sharply the relationship has changed.

OpenAI once promoted its work with Apple as a partnership built around safety, innovation and access to advanced AI. It now describes the lawsuit as aggressive, inaccurate and unnecessary.

Close-up of a metallic knob and translucent buttons on a sleek OpenAI device. Text on the device reads, “OpenAI | 2026.” The design is modern, with a soft, neutral color palette and clear, blue-accented buttons—echoing the refined sensibilities often associated with Jony Ive’s hardware innovations. — Image Credit: OpenAI

From ChatGPT Partnership to Hardware Rivalry

In June 2024, both companies announced an agreement to integrate ChatGPT into experiences across iPhone, iPad and Mac.

Siri could ask for permission to send certain requests to ChatGPT, while Writing Tools could use the model for composition and image generation. The integration gave OpenAI direct exposure inside some of the world’s most widely used personal devices without requiring every user to create an account.

At the time, Sam Altman said the partnership would introduce ChatGPT to users in a new way. Apple presented the external model as an optional extension of Apple Intelligence, keeping permission and privacy controls between the user and the third-party service.

The arrangement made sense while the companies occupied complementary positions. Apple controlled the devices and operating systems; OpenAI supplied a widely recognized generative model.

Consumer hardware changed that balance.

OpenAI is developing a family of AI-native devices with designer Jony Ive, whose work at Apple helped shape products including iPhone, iMac, iPod and Apple Watch. Tang Tan, another longtime hardware figure who previously worked on major Apple products, is now OpenAI’s chief hardware officer.

Once OpenAI began recruiting deeply from Apple’s hardware organization, the relationship was no longer limited to providing a model inside Siri. The companies were moving toward competition over the physical objects through which people may interact with AI.

That competition helps explain why the lawsuit reaches beyond an ordinary employment disagreement. Hardware development depends on manufacturing knowledge, component choices, testing systems, supplier relationships and production methods that can take years to build.

Apple’s complaint alleges that OpenAI benefited from confidential information in those areas. OpenAI argues that the case is being used to slow a potential rival and discourage employees from leaving.

What Apple Alleges Against OpenAI

Apple filed its lawsuit in July against OpenAI, Tang Tan and former electrical engineer Chang Liu.

The complaint alleges that former employees misappropriated trade secrets connected to product development and shared them with OpenAI as it accelerated its consumer-hardware ambitions. Apple has also sought a preliminary injunction intended to prevent the use of disputed information while the case proceeds.

According to Associated Press reporting on the original complaint, Apple accused OpenAI of encouraging recruits to provide confidential knowledge and advising them on how to avoid scrutiny during the transition between companies.

Those are allegations. OpenAI has denied that it requested, received or wants Apple’s trade secrets.

In its public response, OpenAI disputed several details surrounding Apple’s communications and treatment of departing employees. It argued that continued access to some files resulted from Apple’s own account-management practices and said former employees had attempted to handle that access responsibly.

OpenAI also said Tan had instructed his team not to use confidential material belonging to previous employers. Its motion to dismiss asks the court to reject Apple’s case, while its opposition to the requested injunction challenges the evidence and need for immediate restrictions.

Reuters reported that roughly 400 former Apple employees had joined OpenAI’s hardware initiative, according to figures appearing in court filings. Headcount alone does not establish misuse of trade secrets. It does illustrate the scale of the talent movement and why the dispute has become a defining fight over the next generation of personal hardware.

OpenAI’s Device Work Is Continuing

Sam Altman said he does not expect the lawsuit to slow the company’s device program.

That is the most direct business claim in his interview. It suggests that OpenAI has separated disputed material and personnel questions from the work it believes can continue without relying on protected Apple information.

Whether the schedule remains unaffected may depend on the court. A preliminary injunction could impose restrictions before a full trial, depending on its scope. A denial would allow development to proceed without that requested intervention, although the wider lawsuit would remain active.

OpenAI has revealed little about the devices themselves. The company has discussed an AI-native hardware family rather than a single conventional phone replacement, and Altman has spoken about the privacy discomfort created by products that visibly point cameras at other people.

That leaves the design, release timing and commercial strategy largely unknown. No product has been publicly introduced, and Altman’s assurance that development continues does not amount to a launch announcement.

The lawsuit also does not automatically end ChatGPT’s role across Apple platforms. The existing integration remains a separate customer-facing arrangement unless either company announces a change.

For people using ChatGPT through Siri or Writing Tools, the legal conflict currently produces more drama in court filings than difference on the screen. The stranger question is how long two companies can remain software partners while preparing to compete over the device sitting on the table beside that screen.

Image Credit: Samuel Boivin—NurPhoto via Getty Images

The post Sam Altman Calls Himself a ‘Mega Apple Fanboy’ Amid Lawsuit appeared first on AppleMagazine.

Visual Studio Code 1.136 introduces agent merges for pull requests

InfoWorld

5 September 2026 at 00:28

Microsoft has released Visual Studio Code 1.136, an update to the popular programming editor that previews an Agent Merge feature and introduces two experimental features for agentic coding.

VS Code 1.136 was released September 2 and can be downloaded from code.visualstudio.com.

Agent Merge, a preview feature, helps take a pull request across the finish line, Microsoft said. Agent Merge asks an agent to address review feedback, fix failed checks and merge conflicts, and rerun workflows until the pull request is ready to merge. To use Agent Merge, turn on the setting chat.agentMerge.enabled. Then you can enable Agent Merge for a session by running “Enable Agent Merge for Active Session” from the Agents window or selecting the “Agent Merge” button in the title bar.

VS Code 1.136 also introduces multi-root workspaces, an experimental feature that allows developers to work with GitHub Copilot and Claude agents across multiple project folders in the editor window chat view. Agent hooks remain scoped to a single workspace folder, Microsoft said. If hooks are detected in multiple folders, VS Code prompts the user to select the primary folder from which to load them.

Another experimental feature in VS Code 1.136 is chat backgrounds in the Agents window. This allows users to personalize the Agents window with a theme-aware pattern of built-in VS Code icons or an image of their own. Also new for chat are enterprise controls for dictation data. The new controls allow administrators to manage the dictation model and language-model transcript cleanup through enterprise policies. Microsoft said it was possible to require on-device transcription and disable language-model cleanup, so dictation remains available while preventing dictation data from being sent to cloud transcription or a Copilot model.

SoundSource Support Follow-Up

Michael Tsai

By: Michael Tsai

4 September 2026 at 20:19

Chris Barajas:

To spread the word, we announced SoundSource 6 on our blog and sent a mass mailing to our customer list. We also took a more targeted approach, contacting both existing and potential customers who’d previously reached out to us with specific requests. We went back to those folks to let them know about the changes that would interest them most.

[…]

A real live person writes every email response our support team sends, and that human touch takes time. The process is helped along by text snippets, replies that we write once and then reuse and adapt to answer common questions.

[…]

To decide if an email should be sent to a customer, we checked Help Scout’s conversation history, as well as our own custom app for their order history. For example, if someone requested several features across different emails, we’d pick the one that was most impactful, then delete other drafts to avoid inundating them with multiple messages.

Previously:

Reading view

Downloads

August 26, 2026

References & Resources

You may also be interested in:

Discover More from NASA Earth Science

Subscribe to Earth Observatory Newsletters

Earth Observatory Image of the Day

Earth Science Data

Reach without governance is a liability

What you actually gain by building at home

The three homes

Start with the governed interface

When agents start talking to agents

The incentive problem nobody is pricing in yet

The first agent should be the one you can govern

The part you control

The parts you don’t control

Build with your eyes open

Where traditional frontend observability falls short

Monitor accessible journeys, not just healthy pages

Make accessibility part of release health

If users can’t complete the workflow, the system isn’t healthy

Model routing

Semantic caching

Prompt caching

Prompt discipline

Response constraints

Delphi by day

Claude bootcamp

The feature that never ships

Following the money

Getting comfortable

The biggest gains came from removing handoffs

Governance accelerated adoption

Pick your tools and get good at them

Automate the entire life cycle

Design for the velocity you’re about to create

The real transformation is organizational

Why AI-first development?

How to succeed at AI-first development

Add new skills and roles

Consider training programs to ease the transition

Shift toward agentic workflows

Create new testing and review processes

Start small and proceed with caution

Machine Vision Enters the Manufacturing Curriculum

Houston Brings Training Closer to Production

A Useful Inspection Starts With a Specific Problem

Teaching Machine Vision to Handle Ordinary Variation

Smaller Manufacturers Are Already Applying the Training

Access Without a Large Training Budget

Sam Altman Separates Personal Admiration From the Case

From ChatGPT Partnership to Hardware Rivalry

What Apple Alleges Against OpenAI

OpenAI’s Device Work Is Continuing