We have an abundance of AI capabilities. The question for enterprise leaders is increasingly how to make them work together inside a business: which systems they may touch, how their actions are logged, and how people change the way they work.
The Work Has Shifted - The main essay weighs evidence that AI is creating new jobs against a narrower decline in hiring for young workers in AI-exposed occupations. The Economist estimates roughly one million AI-created jobs in America since mid-2023 versus about 200,000 layoffs attributed to AI. Stanford researchers, meanwhile, find employment among 22-to-25-year-olds in exposed occupations running below the trend suggested by less-exposed peers. Those patterns do not establish that AI caused the decline, and neither ledger settles the long-term balance.
What is changing in practice is the work after the prototype. Making AI useful inside a company requires architecture, hosting, permissions, logging, and owners for the conditions attached to approvals. It also requires redesigning workflows and helping managers and teams use the tools in their actual jobs. That work is slow and often understaffed.
The entry-level question belongs in this picture. A co-op student mapping a finance team's processes and concepts illustrates work that AI projects need but busy employees rarely have time to do. Junior colleagues can contribute to that mapping and process repair while learning from experienced staff. Hiring them deliberately keeps both the work and the knowledge in the business.
The AI Slowdown Gets Backers, a Holdout, and a Lawsuit - Anthropic's Dario Amodei called for coordinated limits on frontier AI development. OpenAI's Sam Altman and Elon Musk publicly backed the direction, while Google DeepMind's Demis Hassabis proposed an independent standards body and Meta distanced itself from coordination. Four private plaintiffs have filed an antitrust suit alleging that such coordination would be illegal. The suit is an allegation at the filing stage, and no enforceable pacing agreement or evaluator regime exists yet.
Nvidia Is Financing Both Sides of the Boom - The Economist describes rapid growth and mounting debt among companies renting out Nvidia chips, alongside roughly $300 billion of Nvidia investments and contingent support for customers. Guarantees that protect compute prices and lease commitments could matter most if demand weakens, when the same slowdown may also pressure cash flows. Buyers of AI capacity should expect compute economics to move in either direction.
Also in this issue:
- The Brief - Frontier AI coordination remains unsettled, younger workers face a hiring squeeze, enterprise implementation is understaffed, Bupa used AI to recover business rules from old code, and Nvidia is supporting its customer base as debt rises.
- In Brief - Bupa and Infosys describe an AI-assisted app rebuild, with timing claims reported by the vendors; Google Cloud and Accenture plan to train up to 1,000 engineers who work on site with clients.
- Interstitials / Overheard - A joke about the folding iPhone Duo and the many alarms it could display.
- Meanwhile... - An AI research system suggested orbital maneuvers for a proposed Alpha Centauri spacecraft. A physicist checked the approach; the supporting paper has not been peer reviewed, and the mission has not launched.
- What I'm Consuming - An interview on AI tutoring that preserves learning, BCG's view of an AI-first COO, and a hands-on comparison of GPT-6 Astra and Claude Fable 5.1.
- After Hours - Ben Macintyre's The Spy and the Traitor, the true story of a KGB officer who spied for MI6.