Ask whether AI makes you more productive and you are asking the wrong question, because a job is not a single thing that AI either speeds up or does not. A job is a bundle of tasks, and AI hits each one differently.1

Think about a marketing manager’s week: writing copy, sitting in meetings, building slides, answering email, digging for a number, analysing a campaign, briefing a junior, reassuring a client. AI is brilliant at some of those, useless at others, and quietly harmful at a few. To understand what it does to your week, you have to break the week into tasks and look at them one at a time.

What does AI genuinely speed up?

Email is the clearest win. In a controlled trial, workers with an AI assistant spent about thirty-one percent less time reading email, roughly fifty minutes a week, measured from the system itself rather than from guesses.2 Meetings shrink too: AI can summarise a missed meeting about four times faster, and one firm saw sixteen percent less meeting time overall.3 Around eighty-five percent of people using an assistant say it helps them reach a decent first draft faster.4 Information search ran about twenty-seven percent quicker in one test.5 Each task, taken alone, shows a gain.

So why doesn’t the week collapse?

Because the gains do not survive the journey from the single task to the whole week. Notice the catches. In the meeting test, the people who let AI summarise captured slightly fewer key points than those who did it themselves.3 In the search test, people found answers faster but were no more likely to find the right answer.5 Faster, and very slightly worse. The savings leak away in steering, prompting, and checking.

What about the hard cases?

Here the spread is the story. The makers of one coding tool found it let developers finish a defined task fifty-six percent faster.6 A study of support staff found a fourteen percent gain.7 And a rigorous 2025 trial found experienced developers nineteen percent slower with AI on familiar tasks, even though they believed they were twenty percent faster.8 These numbers are not contradicting each other by accident. AI helps most on tasks within its competence and on less experienced workers, and helps least, or hurts, on expert work in familiar territory.

How should you read the numbers?

With one rule: the cleanest figures come from the companies selling the tools, so treat those as the rosy end of the range, not the neutral truth.4 And remember that around seventy-seven percent of workers using generative AI say it has added to their workload once you count review and prompt overhead.9

The honest verdict is not a debunking. AI really does make many things faster. But the gain you feel and the gain you keep are two different numbers, and almost everything important happens in the gap between them. Productivity is not something the tool gives you. It is something you keep or lose depending on which tasks you hand over and whether you bank the saving instead of spending it on more output.

Frequently asked questions

Does AI actually make you more productive?

Sometimes, and it depends entirely on the task. Controlled studies show real gains in email, drafting, and routine coding, but no improvement or even losses in others, and most measured gains are smaller than they feel.

Why does AI feel faster than it is?

Because there are two stopwatches: the one in your head that reports how the work felt, and the one on the wall that records what happened. AI tells a flattering story to the first while the second often shows a smaller or negative gain.

Which tasks does AI genuinely speed up?

First drafts, summarising, reformatting, routine information search, and well-defined coding tasks show the clearest gains. High-stakes work, anything you cannot verify, and tasks outside your expertise are where it disappoints or harms.

About the author

Tom Goodwin

Tom Goodwin is the author of Don’t Work Harder, a book about taking the time AI gives back as time rather than more work. He is a co-founder of GAMEPLAN and writes on productivity, technology, and the economics of the working week.

Footnotes

Footnotes

  1. The task-based view of work originates with David Autor, Frank Levy, and Richard Murnane, “The Skill Content of Recent Technological Change,” Quarterly Journal of Economics (2003). ↩

  2. Microsoft WorkLab Copilot research (2024): a 31 percent reduction in time reading email, about 50 minutes per week, measured from system telemetry. ↩

  3. Microsoft WorkLab (2023 to 2024): missed-meeting catch-up nearly four times faster and a 16 percent reduction in meeting time, but slightly fewer key points captured by AI users in the summarisation test. ↩ ↩2

  4. Microsoft WorkLab (2023): 85 percent of early users said AI helped them reach a good first draft faster. These are vendor figures, the optimistic end of the range. ↩ ↩2

  5. Microsoft WorkLab (2023): a 27 percent speed-up on information search with no statistically significant difference in accuracy. ↩ ↩2

  6. Sida Peng, Eirini Kalliamvakou, et al., “The Impact of AI on Developer Productivity: Evidence from GitHub Copilot” (2023): a 55.8 percent reduction in time on a defined coding task. ↩

  7. Erik Brynjolfsson, Danielle Li, and Lindsey Raymond, “Generative AI at Work,” Quarterly Journal of Economics (2023): a 14 percent average increase in issues resolved per hour. ↩

  8. METR, “Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity,” arXiv:2507.09089 (July 2025): AI use increased task completion time by 19 percent despite a felt 20 percent speed-up. ↩

  9. Upwork Research Institute (2025): 77 percent of workers using generative AI said it had added to their workload, citing review and prompt-management overhead. ↩