Has AI at Work Freed Up Your Time, or Just Created Different Work?

What 90+ leaders revealed about the hidden trade behind every productivity gain.

May 03, 2026

In 2001: A Space Odyssey, the Discovery One is largely run by HAL 9000, the onboard AI that handles navigation, life support, and ship operations.

You might expect that to leave the crew with nothing to do. The film shows the opposite.

Dave Bowman and Frank Poole spend their days monitoring, checking, running diagnostics, and second-guessing the system. The machine handles the task. The humans handle the judgment.

That is more or less the AI productivity story in a nutshell. The robot does show up. It drafts your email, writes your first pass, summarizes your meeting.

Then it hands the page back to you and waits for the part it cannot do. You sit down to read what it produced and realize the work has simply relocated.

I asked a wide pool of operators, founders, and CEOs a question: has AI actually freed up your time, or has it just changed the shape of the work?

More than 90 leaders shared their experience. What came back was a clearer picture than the productivity hype machine usually allows for.

Here is what they had to say.

By the Numbers

Of the 90+ leaders who shared their experience, the strong majority described the same arc.

AI compressed the work they expected it to compress. First drafts shrank, research shrank, and repetitive customer responses shrank. The blank page mostly went away.

But around 80% reported that the recovered time got reabsorbed almost immediately, mostly into reviewing, prompting, and quality control.

Only a handful described lasting time savings that translated into less work overall. Most described what one founder called “the trust tax” and what another called a faster treadmill.

The pattern is that AI gave most people a different job rather than a shorter one.

A small but meaningful minority told a different story altogether, where the gains were structural and the leverage compounded. Those voices are the ones to pay close attention to, because the difference between their experience and everyone else’s almost always came down to how they designed the system around the tool.

What the Outside Research Says

The leadership experiences I gathered are not isolated. The macro research is starting to land in the same place.

Let’s start with the optimistic number.

A 2025 report from the London School of Economics and Protiviti pulled responses from nearly 3,000 workers and 240 executives across multiple countries. The headline finding put average AI time savings at 7.5 hours per week per employee.

Translated into money, that comes out to roughly £14,000 (about $18,000 USD) annually in productivity gains for each AI-using worker (LSE, 2025).

A full workday recovered every week. That is the version of the story that makes it into the press release.

Now the smaller print.

Earlier 2024 survey data out of the University of Lausanne, since covered by the University of Auckland Business School, asked a follow-up question: where does the saved time actually go?

The answer was less inspiring.

Roughly a third of managers reported losing more than half of their reclaimed time to low-value activity. Over four in five everyday AI users said at least a quarter of theirs got swallowed the same way (University of Auckland Business School, 2025).

The Federal Reserve Bank of St. Louis put the same problem in macroeconomic terms.

Individual workers using generative AI report knocking 5.4% off their weekly hours. Scale that up to the full U.S. workforce and the actual productivity bump shrinks to 1.1% (Federal Reserve Bank of St. Louis, 2025).

Big at the desk, small in the data.

Where does the rest go?

Researchers at the University of California, Berkeley followed 200 workers at an American tech company for eight months and found something that should make every leader pause. The work got more demanding rather than lighter.

Workers expanded their range of tasks, multitasked more, and pushed past the boundaries of their original roles.

The researchers warned that the pace was unsustainable. Downstream effects like quiet job-creep, mental fatigue, and degraded judgment were starting to surface (IT Pro, 2026, citing Ranganathan and Ye).

A 2026 MIT study filled in the technical layer.

Researchers tested 41 LLMs across more than 11,000 work tasks. AI models produced “minimally sufficient” output (a 7 on a 1 to 9 scale) about 65% of the time.

At the top of the scale, where a 9 meant superior quality, the models cleared that bar less than half the time. Adding more processing didn’t help (Fortune, 2026, citing MIT research).

Translation: most AI output is good enough to ship after editing, but rarely good enough to ship without it.

That gap is where the trust tax lives.

My Takeaway: The macro data and the leadership experiences are pointing at the same underlying dynamic.
AI saves time. The time often gets reabsorbed. The quality ceiling on AI output is consistent enough that human review will stay in the workflow for the foreseeable future.
What the research doesn’t fully capture is the second-order effect.
When AI raises baseline output across an industry, the firms that protect recovered time will look slow next to the firms that don’t. That competitive pressure is the engine driving the faster treadmill, and it is going to take deliberate effort to push back against.
The interesting question for the next few years is whether organizations start measuring the right thing.
If the metric stays “hours saved,” the gains will keep evaporating. If it shifts to “decisions made faster” or “errors caught earlier,” we might actually see the productivity numbers catch up to the hype.

The Trust Tax

The first thing nearly everyone mentioned was reviewing.

Polished output looks finished, and that is the trap. Your eye glides over a smooth sentence and your brain wants to believe the underlying claim.

Catching the wrong number buried in a competent paragraph is harder than catching it in your own messy draft.

Joe Spisak, CEO of Fulfill.com, put it bluntly. “I now spend 90 minutes daily reviewing AI outputs that would’ve taken me 45 minutes to just write myself from scratch.”

Spisak runs a 3PL matching platform, so the stakes of a polished but slightly wrong recommendation are significant. He is clear-eyed about what AI costs him.

“I’m not writing first drafts anymore, I’m teaching an algorithm what ‘good’ looks like for our brand voice.”

The work hasn’t gone away. It just looks nothing like the work he signed up for.

Jason Hennessey, CEO of Hennessey Digital, described the same shift in less personal terms. “What emerged was a new responsibility around prompt design and output control. We spend more time shaping inputs and correcting subtle inaccuracies than expected.”

Hennessey’s framing is the one I have heard most often from agency leaders. Quality didn’t drop, but the bar moved.

Their daily workflow now feels less like executing tasks and more like directing a system.

The trust tax shows up worst in fields where small errors are expensive. Insurance, law, medical, and engineering top that list.

James Shaffer, Managing Director of Insurance Panda, described what happens when you publish bad insurance content. “Giving bad insurance advice gets you sued. So instead of staring at a blank page, my team now spends their day aggressively fact-checking and stripping out robotic fluff.”

Volume went up across his operation. Survival depended on treating every draft as suspect.

My Takeaway: Verification is becoming a discipline of its own.
It used to be invisible quality control wrapped inside the act of creation, because when you wrote your own draft the thinking and the checking happened together. AI separates those two things, and that separation is doing strange things to expertise.
Junior people can produce confident output without yet knowing what wrong looks like. Senior people end up doing more reviewing and less making.
Over the next few years, I think we’re going to see a premium on people who are good at the verification layer specifically. Not just domain experts, but verification specialists who can read AI output the way a good editor reads a manuscript and feel the wrong note before they can articulate why.
That’s a new role hiding in plain sight.

The Faster Treadmill

A subtler problem showed up in these leadership experiences, and it might be the most important one.

Speed creates demand for more speed.

Shane Larrabee, President and Founder of FatLab Web Support, runs a team managing over 200 WordPress sites. His team’s response to AI was so unexpected that they had to call an emergency meeting.

“AI freed up our time so fast that we had to hold an emergency meeting about what to do with it. That’s not a joke.”

Tickets that used to take four to eight hours started getting resolved in under an hour, and clients noticed. They responded faster, then they expected faster.

The feedback cycles tightened until the team was producing at factory speed across dozens of clients all day.

“We literally had to regroup and slow down our response times on purpose. Not to be less efficient, but to protect our team from burning out on their own productivity gains.”

They had to artificially slow down to survive their own efficiency. That sentence should give every operations leader pause.

Larrabee also pointed at the revenue side, which gets ignored in most AI discussions.

“If I can complete a project in half the time, the client is thrilled. But they’re not paying me double. The work is worth the same to them regardless of how long it took me.”

Faster, busier, flat revenue. That outcome is roughly the opposite of the four-day workweek anyone was promised.

Riccardo Spagni, founder and developer tools builder at Riccardo Spagni, builds developer tools full time and described a version of this in code. He spends a meaningful part of his day reading and testing AI-generated code, deciding whether it is actually correct.

“The AI is fast and confident, which means it produces a lot of output, and maybe 80% of it is fine. The problem is that the 20% that’s wrong looks identical to the 80% that’s right.”

He stopped writing boring code and started auditing plausible code. He is not sure the second one takes less time.

My Takeaway: The faster treadmill is a systems problem rather than an individual one.
If everyone in your industry is using AI to respond faster, your customers’ baseline expectation moves with the average. The new normal becomes the new baseline, and then the new baseline becomes the new minimum.
The companies that thrive in this dynamic will be the ones who consciously decide what speed is worth and price accordingly. The ones that don’t will run harder for the same money, with thinner margins and more burnout.
This is the Red Queen problem from Lewis Carroll’s Through the Looking Glass, where the Queen tells Alice that in her country you have to run as fast as you can just to stay in the same place. Evolutionary biologists borrowed the phrase to describe arms races between species, where running faster only keeps you even with everyone else who is also running faster.
AI in a competitive industry works the same way. The speed gain doesn’t accumulate as advantage. It resets the floor for everyone simultaneously.
It’s possible that when fast responses become commodity, deliberate slowness will start signaling premium.
A boutique consultancy that takes three days to respond with a thoughtful answer will look different from one that fires back AI-generated replies in two hours. The slow one will read as more careful, more considered, more expensive in a good way.
Slowness will become a positioning move, the same way that handwritten notes and small-batch anything became luxury markers once their cheap mass-produced versions flooded the market.

The Work Moved Upstream and Downstream

A pattern came up over and over in these leadership experiences: the time relocated rather than vanished.

Xi He, CEO of BoostVision, described it as a two-step shift.

“AI didn’t give me ‘free time.’ It shifted where my time goes.”

Upstream, he spends more time being precise about what he is asking for, because vague inputs produce average outputs and you end up forced to think more clearly before you start. Downstream, he spends more time reviewing, editing, and pressure-testing.

“So instead of eliminating work, AI replaced blank-page effort with framing and editing effort.”

That trade is good in his view, though the volume of work hasn’t gone down. The leverage shows up in something else: getting to better decisions in fewer cycles.

Riken Shah, Founder and CEO of OSP Labs, put numbers on the same shape. His team reduced time on routine documentation by nearly 60%, and improved turnaround on internal reports by about 40%.

Quality went up too, though not because AI is good. Quality went up because AI forced clearer thinking upfront.

“Use AI to start fast, question aggressively, and finish manually.”

What “question aggressively” means in practice is the discipline of pushing back on the first draft instead of accepting it. The leaders who get the most out of AI treat the first output as a starting position, not an answer.

Sahil Agrawal, Founder and Head of Marketing at Qubit Capital, put it directly. “AI freed me from first drafts. Then it buried me in editing.”

He estimates the time savings are 30%, not the 70% he expected. His summary is that AI shifted his work from creation to curation.

That word “curation” came up in dozens of leadership experiences, and it is the right word. The job moved from making things to choosing among things and shaping them.

My Takeaway: Curation has always been valuable, we just stopped paying it visible attention.
When production was expensive, the credit went to whoever could produce. The factory owner, the writer who could fill a page, the photographer with the camera. Editors, museum directors, music scouts, librarians, and anthologists worked behind the scenes while the makers got the spotlight.
That balance flips when production becomes cheap. The premium moves to people who can tell what is worth shipping.
If you are early in your career, this is a strategic question to ask yourself. Are you building production skill or curation skill?
Production skill says “I can make this thing.” Curation skill says “I can tell which version of this thing is the right one to ship and why.”
The second one was already valuable, and it is about to become more so.

The Outliers Who Actually Got Time Back

A handful of leaders described a genuinely different experience. They got time back, lasting time rather than redirected time.

The pattern across them was structural in a way worth studying.

Filip Pesek, CEO of DonnaPro, is the cleanest case.

“The founders who struggle with AI are the ones trying to use it themselves for everything. Prompting, reviewing, correcting, re-prompting. That creates more work.”

Pesek doesn’t build AI workflows himself. He has a team that does, and they built automations that handle lead research before sales calls, prefill onboarding forms from transcripts, route invoices, and prepare HR documents.

Tasks that used to take human time happen in the background before he thinks about them.

“I wake up and my briefings are ready. Client context is assembled. Routine documents are handled. The AI layer is invisible to me most days – which is exactly how it should work.”

Invisible. The most mature technology disappears into the workflow.

Iain M. Banks imagined this in his Culture novels, where the AI Minds that run civilization are so embedded that humans rarely think about them. They are present, just not demanding attention. If you have not read any of them, Consider Phlebas is the natural starting point.

The lesson Pesek points at is that the actual bottleneck sits one layer above AI itself. It is whoever owns the system around the AI.

If that is you and you are also the person making strategic decisions, the prompting and reviewing will eat your strategic capacity. If someone else owns the system layer, you experience the results.

Antoine Rousseaux, CEO of Heraclaw, found a different kind of escape. His AI didn’t replace work he was doing. Instead it eliminated a ritual he didn’t realize was eating two hours every morning: scanning newsletters, Reddit, and X to stay informed.

“That is not work exactly – it is staying informed. But it was definitely labor.”

He built an agent to surface the five to ten things actually worth his attention, and the bigger insight came after.

“AI did not just free up time. It changed what staying informed even means. Before, I would read everything to make sure I did not miss something. Now I trust the system to filter, and I have realized 90% of what I was reading was noise anyway.”

His framing of the lesson points at a useful reframe.

“Do not measure AI productivity by hours saved. Measure by what did I stop doing that was not actually moving the needle anyway.”

Mike Montague, Founder of Avenue9, takes the structural approach a step further. He saved 25 hours a week and doubled his revenue last year, with sustained gains rather than partial ones.

His framing is that he became something he calls a “context engineer.” Not a prompter or a user, but a person whose job is to load AI with the deep context it needs: client interviews, branding guidelines, marketing personas, detailed outlines.

“If you treat AI like a talented assistant who needs good context, direction, and oversight, not a magic eight ball, it is the highest-leverage tool I’ve seen in my twenty-five years of marketing.”

His diagnostic is sharp. “The people who say AI hasn’t saved them time are usually still using it like a better Google. The people who say it’s magic are usually not checking its work.”

My Takeaway: The outliers all did the same thing in different ways. They built a system, then stepped out of it.
Pesek delegated the system layer. Rousseaux automated a ritual he didn’t realize was costing him. Montague invested in context architecture that compounds over time.
None of them used AI as a faster typist. They used it as the engine inside a system they designed.
If you are using AI prompt-by-prompt, you are going to feel busier rather than freer.
The leverage isn’t in the prompt. The leverage is in the system around the prompt: what gets fed in, what gets reviewed, who is responsible for the loop, and what compounds versus what has to be redone every time.
That is a design problem. It is also where I think the next wave of productivity gains lives, in the scaffolding around the models rather than the models themselves.

What the Recovered Time Actually Becomes

For many leaders, the time AI gave back relocated rather than disappeared. It became higher-leverage time.

The work they wished they had been doing all along.

Ben Davis, CEO of The Gents Place, described it well.

“AI hands you back an hour and then quietly introduces you to a new obsession. I started spending that recovered time reviewing outputs, refining prompts, and honestly just thinking more strategically about what we were even trying to say as a brand. That’s not wasted time. It’s actually the work I should have been doing all along but kept deprioritizing.”

Across these leadership experiences, the recovered time consistently flowed into three places. Strategic thinking, client relationships, and judgment-heavy decisions that AI cannot make.

Perla Kfouri, Senior Customer Success Account Manager at Microsoft, gave one of the most pro-AI accounts in the pool.

“The tasks that took me hours to finish in the past, can now be done in seconds. I am triple booked for a meeting, AI summarizes key discussion points, list action items and mention my name if I have a task assigned to me. I am off for vacation, AI can catch me up on important emails and list the ones that require action.”

Her recovered time goes into learning more about her customers, planning projects, and staying current on technology. Less time looking through email, more time supporting customers.

Her view is that AI is a learning curve worth investing in, since “spending few hours learning a new tool that will eventually save a lot of unnecessary efforts does pay off.”

The hours moved upmarket, into work that is harder to automate and more valuable when done well.

My Takeaway: The promise of AI was never really a four-day workweek. That was a science fiction overlay we draped over a different reality.
The actual promise is that the lowest-value parts of knowledge work would get cheaper, and the highest-value parts would get more attention.
The leaders I heard from are mostly seeing exactly that. Less time on prep work, more time on the calls that prep work was for.
The risk is that “more attention on high-value work” becomes “more high-value work crammed into the same hours.” That is the faster treadmill again.
The discipline is to actually let some of the recovered time stay recovered. Some of it should become space.
The kind of breathing room that makes the strategic thinking easier. Without that protection, every gain gets reabsorbed and the cycle starts over.

How to Get More Leverage from AI Without Burning Out

A few practical patterns showed up consistently enough across the 90+ leadership experiences. These are the moves that came up most often from the people who seemed to be getting the most leverage out of their tools.

Treat AI like a junior team member, not a magic answer machine. The word “junior” came up dozens of times. Fast, capable, confident, frequently wrong, and in need of review. That mental model handles most of the trust tax on its own.
Invest in context, not prompts. The leverage is in what you load into the system before you ask it for anything: brand voice docs, past examples, client backgrounds, structured templates. Fast iteration on prompts is usually a sign you’re working at the wrong layer.
Build a verification layer. Treat AI output as a draft until a human signs off. Multiple leaders had a literal “human review before it leaves the building” rule, and the companies that didn’t were the ones with horror stories.
Decide what to do with recovered time before you have it. The leaders who reported sustained gains had decided in advance where the saved time would go: strategy, customer calls, a new project. The ones who didn’t decide watched it get reabsorbed within weeks.
Measure the right thing. Hours saved is a vanity metric. Decisions made faster, mistakes caught earlier, customers retained longer, ideas tested per week. Those are the metrics that map to value.

The pattern that ties these together is that leverage compounds when you invest in the system, and evaporates when you don’t. The tool is the same in both cases.

Final Thoughts

The promise was that the robot would do the work.

The reality is that the robot does part of the work, and the part it doesn’t do has gotten harder, faster, and more important. We didn’t get the four-day workweek. We got a shift in what counts as the work in the first place.

The people who’ll do best from here are the ones who can tell which output is worth shipping, which question is worth asking, and which two hours of their day are actually moving the needle.

That sorting capacity has a name. It’s judgment.

It used to be invisible because it was wrapped inside the act of making. AI just unwrapped it.

Now we get to see what it’s worth.

Writing this newsletter requires a lot of caffeine. If you’d like to support it, you can buy me a coffee here. ☕

Interested in sponsoring this newsletter? Learn more here.

Sources

Bick, Alexander, Adam Blandin, and David J. Deming. “The Impact of Generative AI on Work Productivity.” Federal Reserve Bank of St. Louis, On the Economy Blog, February 2025.

Federal Reserve Bank of St. Louis. “Generative AI, Productivity and the Future of Work.” Open Vault Blog, October 9, 2025.

London School of Economics. “AI Boosts Productivity by the Equivalent of One Workday Per Week, New Report Finds.” LSE News, 2025.

MIT research on large language model performance across 11,000+ workplace tasks. Reported in Fortune, April 3, 2026.

Ranganathan, Aruna, and Xingqi Maggie Ye. Research on AI work intensification at an American tech company. University of California, Berkeley. Reported in IT Pro, February 10, 2026.

University of Auckland Business School. “How Is Your Team Spending the Time Saved by Gen AI?” Centre for Digital Enterprise, April 2025. Citing 2024 University of Lausanne survey data.

Zhanna Hamilton

Discussion about this post

Ready for more?