Part 4: Intra-organizational and non-tech agencies

Back to part 3

In previous parts, I’ve laid out the case a) that EA orgs should outsource tech work to agencies and b) that a donor-funded agency has some unique advantages over other funding models.

An alternative mode: agencies as a branch of existing orgs

Instead of having a separate organisation dedicated to tech work, an alternative would be to just expand the domain of an existing EA organisation. This would probably be a form of de facto donor-funding, and could either be in the form of giving its developers occasional flexibility to work on external projects, or giving them the ‘impartial’ task of working on projects from the EA space at all times. This has pros...

  • It might be easier for an established organisation to bring credibility to the idea, and hence less difficult for them to get funding for it, especially if they already had a somewhat flexible budget and were willing to take a chance on the idea

  • It might be more practical to find someone capable of doing the admin work at an org that has already gone through such growing pains, so the marginal overhead of early hires would probably be lower

  • They might have existing resources to spare, eg spare spaces in an office that the developers could work from if they were based in the location

  • It might be more culturally appealing to be part of a larger organisation than one with just two or three staff

  • If the proto-agency used the bonus-payment incentive described in part 3, it might benefit from having someone relatively impartial to judge how to allocate the extra payments in the inevitable edge cases (eg the person with whom the developer has been working suddenly becoming unavailable near the end of the project)

  • Starting out as part of an existing org would mean you’d get de facto charitable status immediately, if that turns out to matter

And cons...

  • Their goals won’t be perfectly aligned—eg an org that focused strongly on the long-term future might not be willing to donate even comparatively ‘low-value’ developer time on their cause to comparatively high value time on global poverty

  • Diluting the focus of an org might reduce the interest that people with specific cause area preferences had in funding it (a donor-funded agency would still lack cause area focus, at least until the movement was big enough to support cause-specific ones, but would isolate it to a single entity, giving funders more total options)

  • It would make reporting outputs difficult—if part of your org’s output was intermittent software support to other charities that would be hard to compare to your other activities (the agency would again isolate this concern to a single entity)

  • If the devs were supposed to be impartial towards the hosting org it would be difficult to ensure no bias

  • If they were expected to be partial to the hosting org, uncertainty about when hosting-org projects might take over would reduce the planning ability of other orgs receiving or hoping to receive aid

  • It would reduce the flexibility to experiment with different organisational structures—for eg it might be impolitic to give too much flexitime or have too flat a hierarchy if the other staff of the hosting org had less freedom

  • The meta-incentives for fixing these problems are bad. The incentives issue is IMO the biggest thorn for donor-funded agencies, but in the long run they would have a strong incentive to fix that issue or lose their funding. Whereas if the incentives for cooperation were difficult to deal with at any given EA org, the simple path would be to prioritise in-house development and ignore the problem, and cooperation would break down.

  • Another way to look at the last point is that the infrastructure for inter-org cooperation is already in place (lots of EA orgs all who know about and are generally willing to at least talk to each other), so if the meta-incentives were right, EA developers would already be doing a bunch of cross-org work, and this whole sequence would be unnecessary.

The arguments here don’t seem conclusive to me either way—I can imagine either way working better. I think either mode would have the advantages and disadvantages discussed in part 3. It’s worth noting that CEA have been spinning off orgs to shrink their domain in order to avoid diluting what they see as their comparative advantages. But Jack Lewars of One for the World has expressed possible interest in supporting such a project.

It might be possible to combine the benefits by starting out an ‘agency’ as a branch of an org while building development capacity and trust, with the explicit intention of spinning it off into an independent donor-funded organisation once it had established a track record .

Generalising the idea to other skills

I’ve discussed the idea of a tech agency, because that’s the area I have the most insight on and conviction in. But there are many general skills which EA orgs need, which might suffer less from professional development (or perhaps not—I don’t know how much eg marketers gain from peer feedback), but certainly have the allocation inefficiencies and difficulties of hiring skilled staff described in part 1 (perhaps more so than a tech agency, given that other areas have fewer objective measures of skill), and where an agency would have the same misc benefits described in part 2 and those described in part 3 if it was donor-funded. Whether or not tech agencies in particular take off, I could imagine agencies being useful eg for marketing, design (if that were separated from tech), HR, perhaps operations, events management, and research.[1] There’s a spectrum of how much in-house specialisation a role has, where eg the C-suite are at one extreme, and the software devs closer to the other—the position of various other roles is debatable.

In fact, ACE, FHI, CSER, MIRI, perhaps Givewell and others are essentially donor-funded research agencies. This suggests a status quo reversal test: would the EA movement benefit a) from splitting its researchers up so that some worked at AMF, some at Givedirectly, some at CEA, some at CFAR, and many primarily as volunteers for tiny independent orgs? or b) from these orgs redistributing their current holdings to ‘end-user organisations’ (eg top Givewell or ACE charities for animals and welfare, and perhaps Ought—or no-one—for AI safety practices), and then fund further research via payments from those organisations? It’s an imperfect analogy since the case for them has different details, but nonetheless, I think that they shouldn’t, which strengths my intuition for a) agencies and b) donor-funding them.

Conclusion & call for involvement

There remains a lot of uncertainty about the feasibility, desirability and even the nature of an EA agency, whether low-bono or donor-funded. Please let me know if you can help answer any of the open questions (by PM if you prefer). In particular:

  • Software developers: how appealing do you find the idea of working at a low bono vs donor-funded agency vs in-house at an EA org vs sticking with non-EA work? How much difference would it make if you were involved in the prioritisation process at a donor-funded org with a remit to find the highest value tech projects?

  • Grant managers: under what circumstances if any would you consider supporting a donor-funded agency? How comfortable would you be with it doing each of the following: a) work for an EA org that would have counterfactually been granted money for a tech project/​tech staff; b) one-off projects that would otherwise have had to rely on volunteers; c) cross-org projects that benefited multiple orgs but that none would individually have paid for; d) internships or other support for less experienced EA developers looking to develop their skills on high value projects? To what extent would you expect to be involved in the agency’s prioritisation process?

  • EA organisations: under what circumstances if any would you forego hiring in-house if an agency was available as an alternative? Would you consider paying a retainer towards one, and if so, how flexible would you be about them doing an amount of work proportional to your fee vs prioritising other work when it seemed to be higher value? (I would suggest answering the latter question anonymously or via PM, since there will be strong pressure and response bias towards giving the ‘optimal’ answer) Would you consider buying (quadratic?) voting rights if donor-funding was insufficient?

  • Individuals who have or have previously had one-off project ideas: reveal thyselves, so we can attempt a headcount! Would you feel sufficiently more optimistic about getting your ideas implemented if donor-funded developers were potentially available that you’d be likely to propose more such ideas or otherwise contribute more if it meant they were implemented sooner/​more reliably?

  • Legal advisors: how difficult would it be to become a charity in various countries? Would there be a clear advantage to registering in the UK vs the EU vs North America, or elsewhere? Would donation-based funding allow you to avoid VAT even for a nonprofit that wasn’t a charity if it were a registered nonprofit? Would any of these answers change if the organisation it was offering services to was a noncharity-nonprofit, or a for-profit?

  • Other staff (eg marketers, designers, and others mentioned in the previous section): do you feel like the arguments presented in this sequence generalise to your type of work, such that it would make sense for EA orgs to outsource it to an agency?

If anyone wants to answer these or make other comments in private, feel free to PM me.


I’m indebted to Grayden Reece-Smith, Roy Rinberg, Michael Peyton-Jones, Srdjan Miletic, Simon Marshall, Adam Millican, Petr Maslov, Johan Lugthart, Jack Lewars, Tamara Borine, Sebastian Becker, Vaidehi Agarwalla, JP Addison and Scott Adams for encouragement and many helpful comments on this essay. Mistakes, heresies and other bouts of insanity are all mine.

  1. Luke Muehlhauser advocated something similar. ↩︎