AI Productivity: 89% Believe the Hype, Reality Says Otherwise

The Number That Hurts

89% of executives say AI makes them more productive. That's a clean stat, looks great in a deck, makes investors happy. Except Foxit and Sapio Research, presenting at SXSW 2026, did something radical: they ran the math.

Executives estimate they save 4 hours and 36 minutes per week thanks to AI. But they spend 4 hours and 20 minutes checking what the AI produced. Net gain: 16 minutes. Per week. About enough time to grab coffee and scan your inbox.

For employees, it's worse. They estimate saving 3 hours and 36 minutes but spend 3 hours and 50 minutes reviewing and correcting. Result: they're losing 14 minutes per week. The tool meant to save them time is costing them time.

It's Not a Bug, It's a Pattern

Here's the thing: this study isn't an outlier. When you cross-reference sources, the same pattern shows up everywhere, like wallpaper you didn't choose.

The METR study, a randomized controlled trial with open-source developers, measured that AI slowed them down by 19%. The kicker: these same developers were convinced it sped them up by 20%. The gap between perception and reality? Nearly 40 percentage points.

Section AI Consulting surveyed 5,000 workers. Over 40% of executives claim they're saving 8+ hours per week with AI. Among employees, 66% save between zero and two hours. PwC surveyed 4,454 CEOs across 95 countries: 56% are getting "nothing" from AI. And an NBER study of 6,000 CEOs and CFOs shows that for 90% of companies, AI has had zero measurable productivity impact over three years.

$250 billion invested in AI in 2024. Zero statistical impact. It's like filling an Olympic pool and realizing the water level hasn't moved.

Why We're All Lying to Ourselves

The problem has a name: "workslop." It's a term circulating more in tech circles, and it nails what's happening. Workslop is AI-generated content that looks like real work but accomplishes absolutely nothing.

Picture a colleague sending you a 15-page report, perfectly formatted, impeccable bullet points. You read it, fix three weird phrasings, rewrite a conclusion. You feel productive. So do they. Except nobody actually thought through this report. AI produced it, your colleague skimmed it, and you spent 45 minutes polishing something that should've been a three-line email.

According to Platformer data, 40% of workslop comes from colleagues, 16% from management. We're passing AI content around like a rugby ball, and everyone's running without knowing where the goalposts are.

There's an even nastier bias: hierarchical overconfidence. 60% of executives say they're "very confident" in AI's accuracy. Among employees, the ones actually verifying outputs daily? Only 33%. Those who touch the tool least trust it most. It's like giving a restaurant five stars by looking at the menu without tasting the food.

We've Been Here Before

This paradox has a famous ancestor. In 1987, economist Robert Solow dropped a line that became legendary: "You can see the computer age everywhere but in the productivity statistics." That was the Solow Paradox.

Companies were investing massively in computing. Consultants promised huge gains. National productivity numbers stayed flat. It took until the late '90s, over a decade, for gains to actually materialize, and only after complete work process overhauls.

We might be at the same point with AI. Microsoft estimates you need at least 11 weeks to see significant gains. And that's the optimistic floor for a well-integrated tool. Most companies are still bolting AI onto processes designed for humans, like dropping a Formula 1 engine into a Volkswagen Beetle without touching the chassis.

Power Users See Through It

There's a glimmer of hope here, and it comes from where you wouldn't expect. The Foxit study reveals a maturity paradox: the heaviest AI users, the ones who use it daily, are also the ones who value human skills most. 75% of them consider human judgment very important.

It's counterintuitive. You'd think the more you use AI, the more you'd blindly trust it. It's the opposite. The more you practice, the more you see the seams. You know where AI excels, you know where it hallucinates, you know when to take back control.

Real gains exist. AES, an energy company, reduced a 14-day audit process to one hour. But that's a case where AI was integrated into a workflow redesigned from scratch, not slapped on top of existing processes.

The problem is these real gains are drowning in an ocean of workslop and validation loops. And a UC Berkeley study points to a perverse effect: workers using AI feel "busier and less able to disconnect." Tasks go faster, but nobody's leaving earlier. Saved time immediately fills with new work, often workslop.

The Validation Loop From Hell

The real productivity sink is the validation loop. AI generates a document. You review it. You correct it. You regenerate. You re-review. Every cycle feels like progress. In reality, you're spinning.

It's like those video games where you grind for hours: you watch the experience bar climb, you feel productive, but when you look up, you haven't advanced the story. The validation loop is productivity's grind.

And here's the worst part: 68% of executives say AI has already triggered restructuring in their company. We're restructuring based on gains that don't exist. 72% cite reskilling as a high priority. But only 12% of employees say they're "very worried" about their jobs. The perception gap is dizzying, at every level.

What We Can Actually Do

There's an emerging concept worth paying attention to: ROE, Return on Employee. 93% of organizations are already tracking it, according to Foxit. The idea is to measure not the tool's ROI, but the actual value produced per person, AI included.

It's a shift in focus. Instead of asking "how much is AI saving us?", you ask "how much value are we actually creating?" That's a question that forces honesty.

If you use AI daily, here's a simple exercise. For one week, track two things: time spent generating content with AI, and time spent verifying, correcting, and reformulating it. Do the math. The result will probably surprise you.

We're not saying AI is useless. We're saying we're telling ourselves stories about the gains, and as long as we're bolting AI onto old processes without rethinking the work itself, we'll keep saving 16 minutes while believing we're saving four hours.

The Solow Paradox resolved when companies stopped putting computers on desks and started reinventing how they worked. With AI, we're still at the desk stage. Real productivity will come when we stop polishing workslop and start asking the right questions. Not "how can AI do this faster?", but "do we even need to do this at all?"