Companies that rushed headlong into workplace artificial intelligence are now reversing course.

The Guardian reports that Meta has scaled back plans to axe staff and replace them with AI agents after employees spent growing amounts of time cleaning up technical and security errors, and that the company is expanding its managerial ranks again after thinning them in a 2023 efficiency push.

Futurism notes that even top tech firms are cutting AI budgets as workers churn out low-effort output while burning through costly tokens to appear productive.

The Guardian also reports that firms from Microsoft to McKinsey are limiting employee usage, while Palantir, Booz Allen Hamilton and Nvidia are restricting employee use of AI companies over data security concerns.

The surprise is that the companies retreating are the ones leading the AI race.

Microsoft executive vice-president Ryan Roslansky publicly lamented 'AI slop' in his day-to-day work, describing a 'doom loop' in which he summarized an AI-generated document with AI and came away with no new thinking, The Guardian reports.

The Guardian quotes Microsoft chief strategy officer Kathleen Hogan conceding that 'access and usage do not equal transformation,' and that adding agents to a broken process still leaves a broken process.

The Guardian also quotes Society for Human Resource Management CEO Johnny C Taylor Jr saying leaders feel they are 'driving the truck and building it at the same time,' and would have introduced AI in a more thoughtful, slower way if they could rewind.

The evidence of underwhelming returns is stark.

According to The Guardian, an Atlanta Federal Reserve working paper found 90% of executives say AI efforts have yet to translate into productivity gains at their companies, while an Ernst & Young survey found 36% of senior AI executives at firms with at least $1bn in revenue reported an AI incident causing data loss, financial or brand damage, or operational disruption.

The Guardian reports Walmart store and warehouse workers contending with false alerts, nonsensical fulfillment routes and unrealistic AI-set schedules, per Business Insider.

It also reports a University of Pittsburgh study concluding productivity did not rise after companies invested in AI and cut staff, and that more than 112,000 US job cuts announced in 2026 have mentioned AI, per Challenger, Gray & Christmas.

The stakes are high for an economy that has poured billions into AI on the promise of transformation. Futurism notes the irony that even Microsoft, whose business depends on selling AI's promise, is rethinking its internal rollout.

The Guardian reports companies are shifting from one-off webinars toward sustained, role-specific training and pairing early-career engineers with seasoned employees, which Microsoft found boosted productivity.

It also reports Meta told employees AI usage would no longer factor into performance reviews and paused tracking employee computer activity to train AI, per Reuters.

University of Pittsburgh professor Mark Ma told The Guardian that many mistakes could have been avoided and that some managers appear to have learned their lesson, though he noted share prices tend to dip after AI-related layoff announcements.