Companies Rushed Into AI. Now Some Are Pulling Back

Summary

Employers are reassessing rapid AI adoption after workers were left correcting errors, reviewing low-quality output and managing new privacy, security and cost risks. Meta reportedly scaled back plans to eliminate staff and expand AI agents after technical and security problems consumed employee time, while rebuilding some management capacity after earlier cuts. Microsoft executives have warned that AI-generated material can create a “doom loop” in which systems produce and summarize repetitive work without adding new ideas, and Walmart workers have described inaccurate task assignments, unrealistic schedules, false safety alerts and inefficient fulfilment routes. As a result, some companies are restricting employee access, strengthening data controls, restoring experienced roles and replacing one-off training with sustained, role-specific instruction. Microsoft says access and usage alone did not produce transformation and found that adding agents to flawed processes simply preserved the flaws; pairing junior engineers with experienced staff improved productivity. The article cites an Atlanta Federal Reserve working paper finding that 90% of executives said their AI efforts had not yet produced productivity gains, while an Ernst & Young survey found that 36% of senior AI executives at large companies had experienced an AI incident involving data loss, financial damage, brand damage or operational disruption. Employees and managers also report major training gaps, and AI systems can generate plausible-looking mistakes that spread through workflows before anyone notices. A case involving a 3M expert report illustrates the legal and professional risks: a lawyer alleged that ChatGPT helped produce a $90,000 report containing unsupported claims, although the expert and counsel said AI assisted with formatting and drafting rather than replacing professional judgment; 3M was later ordered to pay $61.5m to 24 homeowners. Research from the University of Pittsburgh found no productivity increase after firms invested in AI and cut staff, while more than 112,000 US job-cut announcements in 2026 mentioned AI. The article concludes that some managers are reconsidering speed-first deployment as returns remain uncertain and the costs of inadequate oversight become clearer.