- Feedback, references
- Original Google Doc
- Huw, team lead at 80,000 Hours, July 2026
Four weeks with a team lead at 80,000 Hours
Huw leads the AI-safety headhunting team at 80,000 Hours. We worked together daily for four weeks (and are still working together as of 2026-07-27, this is week 6!).
We do ~5 hours a week, including a Monday week planning call, daily morning calls to decide/reassess what the day’s deep-work blocks are actually for, a short afternoon call to hold them, and a shared map of the week so the plan lives somewhere other than his head. His org funds this as a normal contractor arrangement.
He’d been scoring his own weeks — productivity, enjoyment — since before we met, so there’s a before-and-after most engagements can’t give you. Excluding the kickoff week, his weekly averages moved:
- Productivity: 5.6/10 before → 7.5/10 during.
- Enjoyment of work: 5.2/10 before → 6.75/10 during.
In his words:
“I think there is just a very clear difference in the graph of weekly wellbeing and weekly output before and after. It feels quite easy to point at causation here. This is a very longstanding pattern.”
“I don’t think there’s been a week where I haven’t got to the end of it and felt mostly good about the week.”
The mechanism, in his words:
“A great way to get metacognition is borrow someone else’s.”
“There’s something really valuable about literally borrowing memory. Once I can get a difficult problem into someone else’s context, I can talk around it a bit — and suddenly a different route becomes possible.”
Six weeks in, we’re still working together — now with a monthly retrospective, and a deliberate plan to need fewer hours as his own systems improve.
Shared with Huw’s approval. He read this page and signed off on it, including being named and identified by role and organisation, on 27 July 2026.