Knowing when to invest in people, and how much

By gergo @ 2026-08-14T12:16 (+8)

Crossposted on Substack.

TLDR.:

Fieldbuilding programs should balance how much they invest in people based on three main factors: value alignment (shared big-picture goals), skill level and context. I call the right level of investment the “calibration line”, meaning that resources match the individual’s potential for impact at a given point in time.

Programs can go wrong by overinvesting (e.g., giving intensive support to poorly aligned, insufficiently skilled individuals) or underinvesting (e.g., neglecting high-potential people).

Subscribe to Building Capacity

Explaining value alignment, context and skills

Fieldbuilding interventions aim to help people build skills and context to facilitate their transition to high-impact work.

I’m simplifying here, but let’s say the three main characteristics that such programs select for are:

Grouping interventions

Overinvesting:

Pretty self-explanatory: when one organises a program that is too resource-intensive for its target audience. I will caveat here that depending on how much funding you have, it may make sense for you to overinvest in the participants of your program.[3] Most of the impact will come from enabling the top 1-5% of your graduates, but you won’t always know who those people are in advance. There is a trade-off between scaling fast and keeping the same level of cost-effectiveness too.

  1. 12-week AI Safety course for a mid-career professional who hasn’t interacted with AI Safety before
    1. Running a course can be costly, especially when facilitators are paid. My intuition is that it makes more sense for these individuals to first take a shorter introductory program, and then self-select into a longer one afterwards if they remain interested.
  2. 1-year full-time fellowship for a senior academic to explore AI governance research who hasn’t interacted with AI Safety before
    1. It’s tempting to try to attract senior researchers to the field. However, academics are pulled in many different directions — existing research agendas, publishing incentives, collaborators, institutional politics — and changing research focus late in one’s career is not something that’s easy to do.
    2. I heard of one case where an academic was offered a year-long scholarship to research AI governance, but ultimately treated it more as a temporary side project and returned to their original research direction afterwards.
  3. One-off career advice for students who sign up for a call through active outreach, such as via a digital marketing campaign
    1. Contributing to high-impact causes will require engaging more deeply with the ideas anyway, usually through some form of introductory program. Given that, it often makes more sense to focus effort on getting them into the course first, and then speak with those who continue engaging afterwards.
  4. Retreats organised for the local community
    1. See my full reasoning for this in this short post.

The Calibration Line

This is roughly where you want to be: where the level of investment is justified by the participant’s skills, alignment, and likelihood of contributing to high-impact work. Many well-known fieldbuilding programs fit here, which is part of why they continue to get funding.

  1. EAG(x) conference organised by a national EA group
    1. These events are mostly aimed at people who have already engaged with EA and are reasonably value-aligned, which justifies a moderate level of investment per attendee. Occasionally very new people attend as well, but these tend to be experienced professionals whose existing skills partially offset their lack of context.
  2. Career coaching for a student who reaches out after extensively reading 80,000 Hours
    1. Unlike someone reached through active outreach, this person has already self-selected through engaging with EA materials and proactively signing up for a call.
  3. 8-week career planning course for mid-career professionals who want to work in EA
    1. There is already sufficient alignment and commitment to justify a more resource-intensive intervention, including paid facilitators. These participants also tend to have transferable skills that can make career transitions more valuable compared to junior talent.
  4. Regular coaching for an experienced professional transitioning into high impact roles
    1. This is roughly what Successif does for AI Safety. The investment is justified because experienced professionals are the hires the ecosystem most needs right now.
  5. 4-week EA course for a university student who wants to learn more after hearing a fellowship pitch
    1. This is close to the minimum meaningful intervention that can still help someone build context and begin seriously considering a high-impact career path.
  6. 3-month, full-time, highly selective research program with close mentorship for people who already have some background in AI Safety
    1. Programs like MATS fall roughly into this category.[4] These interventions are extremely resource-intensive, but the level of selectivity and mentorship can justify the investment.

Underinvesting:

This is the low-hanging fruit we miss out on, due to prioritising incorrectly or making the wrong assumptions. Promising interventions that are yet to be founded also fall here.

  1. Forgetting about retention: not checking on people whom you have already helped transition their careers, wrongly assuming that all of them are happy where they are, or couldn’t benefit from further connections
  2. Most hubs don’t have AIS city-group organisers working at a professional capacity
  3. Key places where the EA/AIS university groups went dormant or have never been founded
  4. Promising people who are overlooked by selective research programs
  5. Fieldbuilders failing to notice or not having enough resources to accelerate particularly promising members’ careers
  6. There are some people helping newcomers to orient in AI Safety, but there aren't many of them, and they are not well-advertised
  7. Someone receiving poor-quality career advice as the advisor is not experienced enough for the advisee’s skill and context-level

EA’s success story

I want to highlight High Impact Professionals here, as I think they are doing great work, and I think they have gotten the kind of investment-to-value alignment calibration that I’m talking about in this post exactly right. This is because:

You can learn more about their program design here.

Closing thoughts

In my opinion, the kind of miscalibration I'm describing here is one of the easier things to figure out, as some of it can be reasoned through in advance. The harder harder problem seems to be figuring out how to support people once you've decided to invest in them a certain amount. I hope this is useful for anyone currently designing or reviewing a program.

  1. ^

    I’m not saying that they need to agree with every tiny detail about threat models. I’m referring to the big picture stuff, such as “AI will be transformative and might cause enormous harm”.

  2. ^

    An example of a high context individual who disagrees with many takes of the AIS community is Matthew Barnett, e.g. see here.

  3. ^

    Although there is still the opportunity cost with your time.

  4. ^

    Though I think these days many do admit newcomers as long as they are sufficiently skilled.

  5. ^

    That’s not to say that they will quit their day-to-day job immediately. HIP has them build up to action gradually.

  6. ^

    They don't do job placements. In my view, placements carry more risk for senior hires than junior ones, and they typically require the fieldbuilding organisation to cover the individual's stipend for the first few months while the host organisation decides whether to bring them on permanently.