Three things I could do for AI safety field-builders

Everything after someone lands on your site. Do they apply? Do the right ones get through? Do they start, finish, and go on to do something useful?

Week one, the reviewI go through the whole process as an applicant, from your homepage to your decision email. I map the tools and hand-offs behind it, find where people drop out, and talk to a few recent applicants and whoever runs it. You get a written review with recommendations, ranked by effort and impact.

Then, if you want it, version 2I build the improved journey in the tools you already use, with measurement in place. One to two weeks. Working, not a prototype.

The result: a smoother, more efficient application process, where the right people get through and your team spends less time getting them there.

Two weeks to see where you are: what each team actually uses, what’s allowed, what people trust it with, where the hours go. You get a short list of what’s worth doing, in order.

Then we do the first few. Some are quick wins. One or two are workflows worth building properly: an owner on your team, a baseline before we start, a check afterwards on whether the output changed. And working rules short enough that people follow them.

The result: more of your team’s time on the work only they can do, and a number for how much more of it goes to the outcomes in your theory of change.

If you can’t compare one cohort with the next, you can’t say what worked.

Over four to six weeks I go through each programme with the people who run it and find where it pinches: manual stages, stalls, data entered twice, a curriculum that can’t keep up. Then I look for what should be built once and shared: applications, selection, cohort management, delivery, facilitator tools, and a measurement spine.

Start with the spineAt Lightful we asked the same questions at the start, middle and end of every cohort, and across cohorts. That’s what let us show funders what had changed, rather than tell a new story each time. Include the people you didn’t accept and you have something to compare against.

You get a plan: what to fix in each programme, what to share, in what order. If you want, I build the first piece.

The result: a theory of change backed by proof points, so you can show funders what your programmes caused, and grow the ones that work.