Learning from Matching Co-Founders for an AI-Alignment Startup

At EAG London in April 2022, Beth Barnes was looking for potential founders of a new startup that could provide better human data for AI Alignment research. Six of the people interested organised a self-run multi-step co-founder matching process over 7 weeks. It resulted in identifying 2 potential co-founders that have been exploring the founding opportunity in detail.

In this post, we, as part of the group, will describe the process of the co-founder matching and the learnings based on a survey of participants.

The initial idea

The Future Fund’s Project Ideas Competition this March awarded prizes to Marc-Everin Carauleanu and Beth Barnes for suggestions regarding human data for AI Alignment. Beth met with several people at EAG London in April and provided a five-page Google Doc writeup after the conference to everyone who had expressed interest and had spoken with her. The email doubled as an introduction.

Coordinating the process

As Beth implicitly had handed over the idea to the group to self-coordinate, I (Patrick), as one of the participants, set up a Google Form to gauge the level of interest. Everyone was able to see the results. Additionally, I set up a channel in the EA Entrepreneur Slack for further discussion. Based on the interest expressed as well as the time zones of the participants, I scheduled a group call.

We decided to ask Charity Entrepreneurship (CE) if they could facilitate the process as they have several years of experience in finding and matching co-founders for new EA startups. In a call, Joey Savoie provided helpful guidance, sharing documents and later also inviting the team to use the CE offices for a joint weekend at the end of the process. Although they didn’t have the capacity to facilitate the process, their resources were very helpful for us in setting up our own process.

Steps in the matching process

With the input of CE, a group of six people decided to start a matching process that would end in an in-person meeting in June, where the final decision on the co-founders would be made. We think the process length would have been excessive had many participants not been available before mid-June.

Questions for co-founders

The group took the document 50 Questions to Explore with a Potential Co-Founder, removing some and adding additional questions. Once everybody had commented on the document, it was finalized, and everyone filled it out for themselves and shared it.

Discussing questions for co-founders in 1:1s

All participants then set up 1:1 calls with each of the other participants where they discussed the potential compatibility and roles each could take on in the new organisation. No co-founder pair ruled itself out after this process.

5 weeks of online work

In order to bridge the time until an in-person meeting was feasible, a process was set up where each participant worked with one of the other five for 45 minutes each week. As we didn’t have any assignments prepared, we came up with a process where each pair would create an assignment that another pair would have to work on in the following week.

This is how the agenda looked:

WeekGoalWorkstream 1Workstream 2Workstream 3
1Create assignment for week 2Marc, MarisMatt, PatrickRudolf, Simeon
2Work on assignmentPatrick, RudolfMaris, SimeonMarc, Matt
3Create assignment for week 4Marc, SimeonMatt, RudolfMaris, Patrick
4Work on assignmentMaris, RudolfMarc, PatrickSimeon, Matt
5Work on agenda for London meetingSimeon, PatrickMaris, MattMarc, Rudolf


The different workstreams ensured that each pair generated or received assignments from another pair. We also found that trying to quickly brainstorm tasks that are good for testing co-founder fit was itself a good test of fit and forced us to spend solid time thinking about what makes for a good co-founder fit.

Weekend workshop in London

Originally, the idea was to spend several days coworking, however previous commitments reduced this to a two-day weekend. We settled on the UK as one participant was in Cambridge and the other in Oxford, whereas the rest were dispersed in Europe, Berkeley and the Bahamas. London was the easiest to reach for everyone, and Charity Entrepreneurship generously offered us to use their space.

As part of the online working process, we had proposals for the agenda but had to finalize them on the first morning. One of the participants was being held up by a delayed flight connection and would only join in the middle of the second day. In the first session, two of the group expressed that they weren’t considering co-founding at the moment but would offer to join as advisors or help in other ways. One participant wanted to pursue an adjacent but different startup idea and wanted to convince the others to also consider this alternative plan.

We settled on a mix of group sessions and 1:1s where we clarified questions around the organisational setup as well as interpersonal fit. We also decided to write up our current decision concerning who we would like to co-found with most until the following day.

After spending the day together this way and going out for dinner the next morning started with the first reveal. Everyone copied their write-ups and pasted them into a new Google Doc at the same time. This gave us a first view of the current thinking of the others, although, with one missing, it was still a partial view. We then continued the 1:1s until all six were together.

For the afternoon, we agreed that we wouldn’t leave the workshop without having settled on a co-founder pair. After giving the sixth participant some time for 1:1s, we settled on a time for the final reveal of preferences. Concerning the alternative startup idea, there was some discussion around spending some days exploring this, but ultimately, it was decided to go with the existing idea and make a decision the same day.

The reveal showed that three people were still interested in co-founding, with two preferring the same person as a co-founder. This person then held additional 1:1s until he decided who to select. The selection was mainly based on having worked together with the person previously and knowing each other. Both then agreed on a timeline for in-depth research into a product-market fit by doing customer interviews.

The two days ended in good spirits, with everyone joining for dinner and celebrating the successful workshop.

vParticipants from left to right: Marc-Everin Carauleanu, Rudolf Laine*, Siméon Campos, Mathieu Putz*, Maris Sala and Patrick Gruban (*selected co-founders)
Goofing around between serious discussions.

Summary and Learnings

After the process, we set up a questionnaire and surveyed all of the participants with a mix of closed and open questions. The detailed answers are in the appendix.

While all participants rated the results as good, the overall process was rated lower. Most participants didn’t think that an independent facilitator would have helped, and they would still have participated in the same process. Some comments noted that a shorter timeframe would have been preferred.

Of the individual interventions, the 1:1s in London and the narrowing down of the co-founders after the last reveal of preferences of the group were rated the highest (averaging 4,3 and 4,4 out of 5 points, respectively). A few comments reflected the view that meeting in person was important. The prior interventions got mixed reviews (2,3-3,8 points), with the questionnaire for co-founders rated worst (2,3). The five weeks of online work were seen as useful by 4 out of 6 with comments reflecting that it would have been useful to work on specific startup projects like customer interviews. As a general benefit of the process, comments mentioned getting to know more entrepreneurial-minded EAs and being able to spend time together.

As an alternative to this process, one commentator suggested people just informally contact each other and work together. In this case, this might have led to AI alignment researchers getting contacted by several teams in parallel for customer interviews. This could, however, also be solved by coordination without additional interventions.

Another alternative could be to have a process that more closely mirrors CE’s but then would involve someone designing and managing it:

  • Research the product idea in more depth before looking for co-founders

  • Screening potential entrepreneurs before accepting them into the program

  • Designing a program where entrepreneurs work on the company together and thereby getting to know each other

In the way the idea was presented to us and how the group came together, this seemed unusual, so deriving general recommendations might not be useful. As a group, we thought spending time together in person in one place was most useful, so this might be something potential co-founders and supporters could prioritize.

Eventually, the team pursuing this idea decided not to go through with it after spending about a month and a half on it (mainly on customer interviews with alignment researchers).

Personal Takes

Simeon

I think the big mistake that we made was to make the cofounder matching process without having done almost any customer interviews and research on the topic before. I think that it was a significant mistake because it meant that most people had a very fuzzy view of what they were talking about. I personally realized this during the weekend brainstorming sessions with Marc and Patrick, where we did some tasks that led us to have a better view of what were the challenges of the human data field and the potential for improvement.

Thus, if I had to redo it again with a similar setup (i.e. someone comes up with a “startup idea” that hasn’t been tested and hands it over to other people), I would first commission someone (either someone external to the co-founder group or some people of the co-founder group) to do some customer interviews to have a clear view of what’s needed to be successful in this area and what it would be like to co-found something in this area.

Maris

Eventually, it came down to who knew each other most from working together, so ideally, instead of doing test-1-on-1s, we could have done some specific projects over the course of a few weeks to see if we actually fit together. That could have been more useful than doing the 50 questions and the 1-on-1s—these were mainly an introduction, and during the weekend, when we discussed more practically work culture fit-related questions, it seemed like we could have chosen specific questions to zoom in on, instead of doing all 50.

Acknowledgements

We would like to thank the Long-Term Future Fund for approving the short-term request for travel funding and Joey Savoie/​Charity Entrepreneurship for the support and hosting us in London. Thank you to the participants in the co-founder matching process for providing feedback and agreeing to it being openly shareable: Siméon Campos, Marc-Everin Carauleanu, Patrick Gruban, Rudolf Laine, Mathieu Putz and Maris Sala.

Appendix: Post-Workshop Survey

Overall Feedback

We first asked some general questions on a scale from 1 to 5. The tables below show the numbers of results per answer:

Scale: 1 (very bad) − 5 (very good)1 2345
How would you rate the result of the process? 6
Scale: 1 (very unlikely) − 5 (very likely)12345
If you would have known how the process would have been (without knowing the result of the weekend), would you have still participated? 132
Do you think an independent facilitator would have been good in the process? 132

Question: What would you change in the process?

Answers

Seemed like it was both too fast and too slow. Eventually it came down to who knew each other most from working together, so ideally instead of doing test-1-on-1s we could have done some specific projects over the course of a few weeks to see if we actually fit together. That could have been more useful than doing the 50 questions and the 1-on-1s—these were mainly an introduction, and during the weekend when we discussed more practically about work culture fit related questions, it seemed like we could have chosen specific questions to zoom in on, instead of doing all 50. But that might have been an effect of having done the 50 questions and realising that a lot of the answers simply align and are thus not as useful.

More evidence-based strategy

I would probably not want a formal process at all. Just people who are interested in working with each other informally contacting each other.

I would do some customer interviews before the final weekend + I would ensure that everyone has a better inside view of what the mission of the startup looks like.

  • Try to come up with better questions for the 1:1s

  • Try to narrow down earlier who would be really be interested in co-founding

  • Try to work on actual startup tasks

Question: What did you like best or find most helpful?

Answers

It was good that there was some process to it, I actually found the 50 questions useful in terms of orienting myself towards the process, but any short overview of what the process would looked like would have sufficed. I liked the in-person weekend in London and how we ended up approaching it in a relaxed way—decided the plan once we were together. It seemed quite useful.

Getting to know other entrepreneurial EAs

Knowing who’s interested.

The in-person 1o1s were very useful. The fact of thinking about the company was also very useful. We maybe should have done more of that before the final week-end.

Spending time together in London and getting to know people more

Feedback on the interventions

Question: Please rate these elements of the process according to how useful you found them

Answers

Scale: 1 (not helpful) − 5 (very helpful)12345
The whole process 123
The initial questionnaire where you expressed interest 2 3
The initial group call 14
The questionnaire of 50 questions for co-founders1311
The initial 1:1s where we talked about the 50 questions 141
The Slack channel 1221
The 4-weeks process where we had 1:1s and created questions/​answers11 4
In London: The whole weekend 1113
In London: The 1:1s 42
In London: The initial reveal Sunday morning 212
In London: The narrowing down of founders after 4pm 32

Question: Anything else you would like to add?

Answers

I think I learned a lot from the process, also on a meta-level. I’ve never co-founded an organisation like this before and it was an interesting test of finding out which traits make for a co-founder that people like to pick as their number one, what kind of questions are useful to ask and when, how important it is to spend time in the physical space together, and whether I think I’m personally a good fit to found a company. Very grateful for the experience <3

  • If it had not been the case that most people couldn’t start until the end of the process anyways, I would’ve considered the process way too long. I think if running something similar with people who are ready to start sooner, the process should definitely wrap up in a tighter time schedule (perhaps by being shorter but more intense or just by having fewer hours).

  • Figuring out who you can work with on ambitious projects is enormously valuable. Having a group of competent people who you know how to work with /​ how well you’d work with is enormously valuable.

  • Differentiating among a group of people who are already competent impact-oriented EAs is quite difficult without either lots of time with someone, or experiences that mirror as closely as possible the actual process of doing work of the type that the project would involve.

  • It seems surprisingly difficult to make artificial tasks that are useful work tests, especially when a big part of the work that the task is supposed to test for is open-ended agentic execution toward complex goals in the real world.

  • Getting together in person to hash things out (rather than relying on scheduled remote calls) remains surprisingly effective.

No comments.