AI Productivity Gains Are Hiding Employee Burnout, Engineers Warn

Three of a team's strongest mid-level engineers resigned within two months of a 40% speedup in output — because AI productivity gains were quietly burning them out. Sreenivasa Reddy, a lead software engineer at AT&T, described the case: his team's adoption of AI looked textbook at first, until its mid-level engineers, buried in reviewing and fixing AI-generated code with no metric or review credit for that work, walked out the door.

Illia Smoliienko, founder and engineering leader at Sivense with over 20 years in tech and a decade in leadership, argues this is becoming a structural problem. "With AI tools taking over part of the workload, people can still hit their usual output even when they are a step away from burnout," he wrote. His old early-warning system — an unscheduled one-to-one whenever someone's delivery slowed — no longer works, because AI masks the slowdown.

What does the research show?

The pattern is measurable. A study of 442 employees at tech companies, published in Frontiers in Psychology, found the link between burnout and a drop in productivity was weaker among heavier users of AI. The dashboards stay green while the people behind them run dry.

Researchers at the University of Lausanne surveyed 83 managers and found that 82% save an average of 2 hours and 50 minutes per week thanks to generative AI — about 1.5 working days a month. Where does that time go? More than half of the managers spent it on getting more work done. 36% didn't even realize AI had saved them time.

Workday surveyed 3,200 managers and employees across three regions: 85% said AI saves them between 1 and 7 hours a week, but around 37% of that saved time goes to fixing and rewriting what AI has generated. This correction work, Smoliienko notes, often falls to middle-level roles — exactly the pattern in Reddy's AT&T team.

How does AI reset expectations?

AI does not just absorb routine tasks. It gradually raises the bar for what counts as a normal workload and complexity level for a given role.

An architect on Smoliienko's team now expects junior developers to be more independent: they should work questions out with the AI assistant first, then return with a specific problem and possible solutions rather than expecting a lecture on basics. In a Harvard Business Review article, the head of workforce innovation at KPMG said she now expects associates to come to meetings as well prepared as managers once did, because AI makes that possible.

For juniors, that can accelerate growth. But the productivity boost is not spread evenly across a team — and that unevenness has consequences, usually landing on the middle.

How can managers spot burnout behind high metrics?

A drop in output used to be the signal to check in. Now it may come too late or never appear at all. Smoliienko offers three starting points for leaders:

• Acknowledge and value "invisible" work. Reviewing generated content and fixing mistakes barely shows up in productivity stats. The people doing it may have fewer completed tasks to their name, yet they keep the whole team's numbers high. This work should count in workload balancing and performance reviews.
• Don't turn every saved hour into more work. When back-to-back task completion becomes the norm, the team operates on its limit. Decide in advance whether freed-up time goes to more work, quality control, learning, or recharging.
• Don't confuse using AI with efficiency. The test is whether the tool helped make a better decision. Otherwise, AI adoption becomes another metric people optimize for the sake of optimizing.

Smoliienko says he no longer takes strong numbers as proof a team is doing fine. After asking "How much have we achieved?" he now asks one more question: "What did this result cost the team?" For companies scaling AI, that second question may be the one that determines whether the productivity gains — and the people producing them — survive.