Empirical evidence shows AI dependency degrades professional skills: A Nature-featured study found that physicians’ adenoma detection rate during colonoscopies dropped from 28.4% to 22.4% when AI assistance was unavailable, after they had grown accustomed to using it. Similar patterns are emerging among software engineers. Founders and operators building AI tools must intentionally design for skill retention, not just task automation. Nature
Enterprises are pulling back on AI as costs outstrip ROI: The FT reports that even companies that conducted mass layoffs are finding AI unjustifiable on cost grounds for most use cases. None of the major AI companies are profitable — they are still charging below their actual delivery cost. For teams building enterprise AI products, proving concrete value-per-dollar is about to get much harder. Financial Times
Miami startup Subquadratic releases third-party benchmarks for its transformer-alternative architecture: The SubQ model claims to have broken through a key bottleneck holding back LLMs, offering 12x context processing, lower cost, and lower energy use while roughly matching top models from OpenAI, Anthropic, and DeepMind on coding tasks. Independent evaluation by Appen partially validates the claims, but the model is not yet publicly available. MIT Technology Review
Sundial launches private beta for agent-native collaborative editing: A workspace where humans and AI agents (Claude Code, Codex) edit documents, markdown, and LaTeX files together in real time. Every edit is tracked to the agent turn that produced it, and humans retain final sign-off. Directly addresses the review bottleneck of the agent era. Sundial
OpenAI’s head of enterprise AI sales, Barret Zoph, departs after five months: Zoph rejoined OpenAI in January after co-founding Mira Murati’s competitor Thinking Machines Lab. His second exit from the company raises questions about internal stability and strategic alignment in OpenAI’s enterprise push. The Verge
Amazon drops nearly finished Sam Altman biopic after $50B OpenAI-AWS deal: Luca Guadagnino’s “Artificial,” starring Andrew Garfield as Altman, is being shopped to other studios after Amazon MGM pulled distribution. A case study in Big Tech conflicts between cloud partnerships and content arms. Variety · The Independent
Norway imposes a near-ban on AI in elementary schools: The first major Western nation to severely restrict AI in the classroom. Edtech startups and teams building AI learning tools should anticipate rapidly diverging regulatory landscapes across jurisdictions. Reuters
AI-induced “deskilling” is now a measured phenomenon in both medicine and software engineering: This is no longer a speculative concern — it is quantifiable. Teams designing AI tools need to ask at the design stage: is this tool augmenting human judgment or replacing it? Nature
Growing demand for AI output review and verification tools: The Nature deskilling study and Sundial’s approach point in the same direction — the more output AI generates, the more valuable tools become that track and assure what humans have actually reviewed. Code review, content verification, and decision audit trails are fertile ground. Nature · Sundial
AI cost-efficiency stacks are the next big opportunity: The FT-reported enterprise pullback on AI costs is a near-term friction that creates demand. Tools and services that enable cheaper fine-tuning, smaller-model migration, on-device inference, and exploration of novel architectures like Subquadratic’s will find willing buyers. Financial Times · MIT Technology Review
Is Subquadratic’s SubQ a genuine breakthrough or AI’s Theranos moment?: Appen’s third-party validation lends credibility, but without public model access, independent reproduction is impossible. Watch for broader model availability and additional independent benchmarks. MIT Technology Review
Long-term organizational impact of AI deskilling: Current evidence is limited to medicine and software engineering. Whether this extends across all knowledge work, and whether training methodologies exist to maintain skills alongside AI use, remains an open question. Nature