Companies rushed in on AI. Now some are pulling back

Companies that raced to deploy generative AI are now reining in use after widespread practical problems undermined promised productivity gains. Big firms including Meta and Microsoft have scaled back or restricted internal AI projects because employees spent increasing time fixing low-quality, AI-generated outputs, and because tools introduced security and confidentiality risks. Surveys and studies back up the pullback: an Atlanta Fed working paper found most executives report no productivity lift so far; an Ernst & Young survey said more than a third of senior AI executives experienced incidents causing data loss or operational damage; and a University of Pittsburgh study found no productivity increase after firms invested in AI and cut staff. Many companies that embraced rapid rollout are shifting to tighter controls, role-specific training, and restoring managerial capacity.

The practical failures are concrete and varied. Workers report an “ocean of sameness” from AI drafts and bots that create unrealistic schedules and faulty routing at retailers, while legal and technical work contaminated by AI-assisted drafting has led to courtroom exposure and costly verdicts. Employers are discovering gaps in training and oversight - managers often feel outskilled and employees say training is inadequate - so firms are pairing junior staff with seasoned employees, pausing practices like using employee telemetry to train models, and limiting engineer and consultant access to external AI services to contain risk and costs.