Dr Henry Shevlin is Associate Director of the Leverhulme Centre for the Future of Intelligence at Cambridge, co-founded the UK’s first master’s programmes in AI ethics, and has just been hired by Google DeepMind to work on machine consciousness and AGI readiness. On Monday 7 September he posted this:
Daily reminder: the world is mostly run by people of middling competence and unusually high agency. If you want to make a difference, start agency-maxxing.
— Henry Shevlin (@dioscuri) September 7, 2026
112,719 views. 4,232 likes. And — the number that says what kind of post this is — 1,093 bookmarks.
This is the suffix doing the job it was built for
Almost everything we’ve logged this year has been the word being used about something. A governor captioning a video. Kim Kardashian on a paddleboard. A researcher describing what language models do to benchmarks. Newsrooms manufacturing a trend out of it.
Shevlin isn’t describing anyone. He’s prescribing. Start agency-maxxing is an instruction, addressed to you, about a thing you should go and do more of — which is precisely what the construction was invented for on the forums a decade ago, and precisely what the last year of mainstream adoption had mostly stopped using it for.
The word went up the respectability ladder by becoming a joke, a caption, a headline generator. Here it arrives at the top still meaning what it originally meant.
And the stat he picked is the interesting part
Every maxxing branch optimises something measurable: a jawline, a batting average, a PSA grade, a country count. Agency has no unit. You cannot screenshot your agency. There is no leaderboard, no before-and-after, no protocol to follow at 6am.
What Shevlin is claiming is that it’s nonetheless the highest-leverage stat available — that competence is roughly evenly distributed and mostly middling, and the actual differentiator among the people running things is how willing they are to just start. Which makes agency-maxxing the meta-branch: the one you’d have to max first in order to max any of the others.
That also explains the bookmarks. A thousand people did not save that post because it was funny. They saved it because it read as a usable instruction, filed next to whatever else they intend to act on and mostly won’t. The irony is doing its own work there and we’ll leave it alone.
Two Cambridge academics, six weeks apart
For the record, this is the second time this summer that a Cambridge researcher has put this suffix into circulation from inside the university. In July, Dr Adam Hunt — evolutionary psychiatry, same city — used “maxxing” as a bare transitive verb in a technical argument about AI scaling, to 614K views.
Different departments, different registers, no connection between them. One used it as precise jargon for machines over-optimising a metric; the other uses it as sincere advice to humans about under-optimising one. Both assumed their audiences would follow without a gloss.
Somewhere there is a forum post from 2016 where someone had to explain what -maxxing meant. It is now a word that Cambridge gives advice in.