Why does AI make employees work harder instead of saving them time?

The Hidden Cost of AI Productivity Gains at Work

Protiviti's August 4, 2026 AI Pulse Survey found only 13% of HR leaders believe their job designs are AI-ready, and ActivTrak's 164,000-worker study shows focused work fell 9% after AI adoption because freed-up time gets reassigned, not returned. Judson Brewer's anxiety habit-loop model from Managing Your Anxiety explains why, and points to the fix: interrupt the reward, not the tool.

Three weeks ago a founder I coach told me her team's average project turnaround had dropped from three weeks to eight days after rolling out Claude across the org. She was thrilled for a month. Then her best engineer asked for two weeks off — not vacation, just recovery — and I realized the eight-day number was hiding the actual story.

That story got a name on August 4, 2026, when Protiviti published its fifth AI Pulse Survey, "The AI-People Conundrum: Learning to Lead, Not Lag." Only 13% of Chief Human Resources Officers strongly agree their organizations' job designs are AI-ready, compared with 28% across the wider C-suite — and just 14% say their learning-and-development function is ready, versus 36% company-wide. Nearly eight in ten executives still expect AI to lift profitability over the next three years. That gap between what leadership expects and what HR knows is actually happening isn't a training-budget issue. It's a workload-redistribution issue nobody is tracking.

The mechanism showed up in a bigger dataset the same week. ActivTrak's analysis of 164,000 workers' digital activity, comparing the 180 days before and after they started using AI, found time on email and messaging more than doubled and business-software use rose 94%, while genuinely focused, uninterrupted work fell 9% for AI users and held flat for everyone else, as the Wall Street Journal reported this month. The study also found a productivity sweet spot: employees who spent 7% to 10% of total work hours actually using AI showed the highest output — but only 3% of AI users land in that range.

None of this means AI isn't saving time. GoTo and Workplace Intelligence's Pulse of Work 2026 survey of 2,500 employees found people save roughly 2.6 hours a day using AI tools. The catch, which Fortune summarized on August 4, 2026: that saved time isn't coming back to the employee. "The time savings went to the company, but the pressure went to the employee," as the piece put it — and 50% of workers in the GoTo data now say they rely on AI too much, with 39% (46% of Gen Z) saying that reliance is making them less intelligent. Forty-three percent admit they've shipped AI output despite suspecting it was low-quality — the "workslop" problem researchers at Stanford and BetterUp named in 2025, still alive a year later.

I kept looking for a management framework built for "we have more time and we're using it to feel worse," and the one that fits isn't a productivity book — it's Managing Your Anxiety, the Harvard Business Review collection built on Judson Brewer's anxiety habit-loop research. Brewer's loop has three steps: a trigger, a behavior, and a reward. Applied to a person, the trigger is an anxious thought, the behavior is worrying, and the reward is a false sense of having "done something." Applied to a team that just adopted AI, the trigger is freed-up capacity, the behavior is assigning more work into that gap, and the reward is a productivity chart that looks great in the board deck. It's the same loop. The organization is worrying its way through a windfall instead of banking it.

Brewer's actual fix for a habit loop is to interrupt the reward, not fight the trigger — and that's the version I now run with teams after any real AI rollout, once a month, for two quarters. I call it the Capacity Audit:

Name the freed hours. Ask each team lead for an honest estimate of hours AI actually recovered that month, not the vendor's claim. Most leaders can't answer this on the first try, which is itself the finding.
Trace where they went. Was each recovered hour reassigned to new output, absorbed into reviewing AI mistakes, or genuinely given back as slack? ActivTrak's 94%-more-software-use number is what "reassigned" looks like in the data.
Cap the reassignment. Before the next sprint, leadership explicitly decides what percentage of recovered time is protected, written down, checked again the following month.
Ask who's in the 3%. Identify people already at the 7%-to-10% AI-usage sweet spot ActivTrak found, and study what they're doing differently before scaling it.

I'll say plainly where this breaks: the Capacity Audit only works if someone with real authority owns step three, because "protect the freed time" is the one line every calendar invite conspires against. I ran this with one portfolio company for a full quarter and watched the protected hours quietly vanish in month two, once a client deadline moved up — the audit caught it, but catching it after the fact didn't undo the burnout already underway. It's also not a fix for teams where AI genuinely isn't saving time yet; if step one comes back near zero, the problem is adoption, not workload design.

The honest takeaway isn't that AI adoption should slow down. It's that treating recovered time as free capacity, rather than capacity needing the same deliberate allocation as headcount, is what's quietly running people into the ground while the productivity number on the dashboard keeps climbing.

Sources: Protiviti, "The AI-People Conundrum: Learning to Lead, Not Lag" (fifth AI Pulse Survey, published August 4, 2026); Fair Play Talks' coverage of the Protiviti survey (August 4, 2026); Fortune, "How AI turned your best work into the bare minimum" (August 4, 2026); Futurism's coverage of ActivTrak's 164,000-worker Wall Street Journal-reported analysis (August 2026); GoTo and Workplace Intelligence, "The Pulse of Work in 2026"; Judson Brewer's anxiety habit-loop research, as presented in Managing Your Anxiety (Harvard Business Review Press).