Microsoft eliminated 4,800 positions on Monday, roughly 2.1% of its global workforce, even as it channels an estimated $190bn into AI infrastructure this year. In a memo to staff, chief people officer Amy Coleman wrote that the eliminated roles “are not being replaced by AI,” but added that “AI is changing how work gets done.” The layoffs, concentrated in commercial sales and Xbox, follow a pattern across Big Tech: Meta, Amazon, and Google have all trimmed headcount while accelerating AI spending. The tension is not confined to Redmond. Uber deployed Anthropic’s Claude Code to 5,000 engineers, achieving 95% adoption and AI participation in 70% of new code, yet exhausted its annual AI budget in four months. The company has struggled to link the surge in token consumption to visible business improvements, a dynamic that financial officers across the industry are now confronting as generative AI’s per-query pricing model replaces traditional per-seat software licenses.

The cost overruns have forced a strategic pivot. Microsoft, OpenAI, and Amazon are each building dedicated deployment units—thousands of engineers embedded inside client firms to redesign processes and integrate AI into operations. The model revives the concept of Forward Deployed Engineers, and it amounts to an admission that the bottleneck is no longer model capability but implementation. In finance, JPMorgan Chase, Goldman Sachs, and Citigroup have moved AI agents into middle-office flows, but all retain mandatory human supervision for critical decisions. Research presented at the CERALE 2026 forum in Rio de Janeiro, analysing 1,573 observations of Argentine sovereign debt, found that AI models excel at identifying when to allocate capital but cannot replace human judgment on when to override the model during structural crises.

Education and mental health are confronting a parallel reckoning. The University of Chicago Law School is requiring laptop-free first-year classes and oral defences of research papers after a wave of AI-assisted cheating scandals at Brown and Harvard. Wealthy families in the United States are shifting enrolment toward schools that blend AI-driven personalised learning with project-based life-skills training, though Stanford researchers caution that rigorous evidence of effectiveness remains absent. In mental health, millions now use general-purpose chatbots for “therapy micro-bursts,” but specialised AI tools struggle to retain users who find switching costs too high. A lawsuit filed against OpenAI in August 2025 over inadequate safeguards underscores the risks of relying on systems that can co-create delusions. Researchers at Wharton, in experiments with 1,372 participants, documented a tendency toward “cognitive surrender”: users accepted AI recommendations 79.8% of the time even when intentionally wrong.

The IMF’s July 2026 World Economic Outlook projects global growth of 3.0% this year, with economies tied to AI supply chains outperforming, but warns that investment bubbles may not lift all boats. Indonesia’s growth is held steady at 5.0%, while China slows to 4.6% and India to 6.4%. The fund’s report notes that AI hysteria is manifesting in real-world investment as well as stock prices, raising the stakes for the coming quarters. The next factual milestone is the October update to the World Economic Outlook, which will reveal whether the spending surge is translating into durable productivity gains or merely inflating a new cost structure that even the largest firms are finding difficult to sustain.