Claim check
That number you heard. We checked it.
Numbers about AI at work travel fast and lose their fine print on the way. Here we take the ones people quote in meetings and decks, read the original source, and grade them in plain words. We grade claims, never people.
Holds up: the source says it. Holds up with a catch: true, but the limit matters. Stretched: the way it is quoted says more than the source. Not supported: the source does not say it.
#003 "95% of AI pilots fail"
Grade: Stretched. Checked 12 November 2026.
The source is a July 2025 report from MIT's Project NANDA, "The GenAI Divide". It says that "95% of organizations are getting zero return". It is based on a review of more than 300 public AI initiatives, interviews with people from 52 organisations, and a survey of 153 senior leaders at four industry conferences. Success meant deployment beyond the pilot with measurable results, judged six months after the pilot.
The catch: the report itself says its figures are "directionally accurate based on individual interviews rather than official company reporting", and that success definitions may differ across organisations. It was not peer reviewed, and others, among them a Wharton professor, publicly questioned how the 95% was reached. So "most companies struggle to show a return yet" is fair. "95% of AI fails" is more than the report can carry.
Source: Challapally, Pease & Raskar, The GenAI Divide: State of AI in Business 2025, MIT NANDA, July 2025.
#002 "AI makes developers faster"
Grade: Stretched, and out of date either way. Checked 12 November 2026.
In a randomised study from METR, 16 experienced open source developers worked on 246 real tasks, each randomly assigned to allow AI or not. With AI they took 19% longer. Before the study they expected AI to speed them up by 24%, and afterwards they still believed it had sped them up by 20%.
The catch: 16 developers, one kind of work, early 2025 tools. In February 2026 METR published newer data and now says the 2025 results no longer reflect the current impact of AI tools. So neither "AI makes you faster" nor "AI makes you slower" holds as a general rule. What does hold: how fast it feels and how fast it is can be different things. Time it.
Source: METR, Measuring the impact of early 2025 AI on experienced open source developer productivity, July 2025, and METR's February 2026 update.
#001 "40% of workers get AI workslop"
Grade: Holds up with a catch. Checked 11 November 2026.
In a survey of 1,150 full time US desk workers, 40% said they had received AI work that looked finished but was not in the last month. Each case took on average 1 hour and 51 minutes to deal with.
The catch: a self reported online survey, by researchers partly linked to a coaching company. Time and money figures are estimates from what people remembered. Good evidence that it is common, not of its exact size in your team.
Source: BetterUp Labs and Stanford Social Media Lab, 2025; Niederhoffer et al., Harvard Business Review, September 2025. Full guide
Heard a number? Send it in
If someone quoted an AI at work number at you and you are not sure about it, email it to [email protected] with where you saw it. We check it against the original source and add it here.