Three things I could do for AI safety field-builders

A hypothesis from conversations at EAGx Oxford. Tell me if it’s wrong.

I talked to people who run the programmes that bring new people into AI safety, and to people who’ve been through them. The same problems came up every time. Here’s what I think would help.

Tell me about your programme

My theory of change: take the friction out of running AI safety field‑building organisations, so they can scale what works.

What I could do

Start where you are.

If your programme is already running, start with the Red Team; if you’re piloting something new, start with the build. Either can lead on to AI uplift, and everything is scoped with you before we start.

Red Team

Red Team your application process

For anyone running a course, fellowship or programme with an application process.

Do the right people get through? From there it follows the whole journey: who applies, who drops out, who finishes and who goes on to do something useful.

In a week I go through it as an applicant, from homepage to decision email, map the tools behind it and talk to recent applicants and your team. You get fixes ranked by effort and impact, and I can build them.

The result: more good applicants through the same team, so the programme can take the growth in applications without adding people or turning good candidates away.

Pilot build

Piloting a new programme or tool

For organisations piloting a new course, fellowship, programme or tool.

You bring the content. I help design everything around it: the applicant journey, the tools, the engagement funnel and the tracking, so the pilot works from day one and tells you what to change before you scale.

The result: a pilot that works from launch, with the evidence to decide what to scale.

Where it can go next

AI uplift

AI uplift for AI safety organisations

For teams already using AI who want the hours it saves, and the judgement calls it touches, to be better.

It starts with the few workflows that matter most and still take the most time. It’s less about adopting AI than using it well and responsibly: where it saves real time, and where it makes calls a person should check. We fix those, with a baseline to show what changed.

It ends with someone on your team who owns it and carries it on.

The result: time back for the work in your theory of change, and a number for how much.

Background

Where this comes from

Twenty years building things in mission-driven organisations. JustGiving Crowdfunding, from nothing to £100m raised. BRIDGE at Lightful, a six-month programme for charity leaders, and the measurement behind it. Lightful’s AI Squad. These days I build with Claude Code and Codex as much as I plan.

On AI safety: BlueDot’s AGI Strategy course and CEA’s career bootcamp, both this year. I’m not a researcher. For most of this year I’ve been the applicant.

How to start

Tell me about your programme.

Whether it’s running now or still a pilot, tell me what it is and where it’s getting stuck, and we can work out together whether any of this fits.

Thanks. I’ve got it and I’ll be in touch within a couple of days.