If AI has done one thing in your organisation over the last few years, it's probably made it really easy to produce more “stuff”. More reports, more slide decks, more summaries, more code. I've been wondering whether most organisations are actually better off because of all the extra output?
This month OpenAI shared 722 draft maths papers in one dump, claiming progress on more than 370 problems human mathematicians haven't been able to crack through history. Which sounds incredible (and is!). But almost none of them have been through peer review yet, and working out which ones are correct and which ones matter could take humans decades. So the humans are now the bottleneck.
In this week's video essay I look at why so much workplace AI ends up producing lots more output (slop, anyone?) but not much ROI.
So I think there are two decisions here that need to be made within organisations. Watch the video to hear them…
Sources & citations
- OpenAI, "Sharing AI progress in mathematics", 6 Oct 2026
- OpenAI on GitHub, openai/math manuscript repository and release history
- Quanta Magazine, "AI Proved the Quasi-Riemann Hypothesis. What Happens Now?", 19 Oct 2026
- Nature, "OpenAI posts 700 maths preprints online", 7 Oct 2026 (paywalled)
- Ethan Mollick on LinkedIn, 10 Oct 2026
- McKinsey, Technology Trends Outlook 2026, Sep 2026
- Deloitte, State of AI in the Enterprise, Jan 2026
