I wrote this for the SaferAI team. I’m publishing it with some small edits, in case it’s useful to other orgs. A note on scope, this post is about sourcing candidates specifically. It doesn’t cover how we define a role before opening it, how we assess candidates once they’re in the process,….
I recently listened to this podcast with Adam Ward, who runs talent at Cursor, and read this post by Sam Altman on hiring. These two resources prompted me to rethink how we approach talent building. Below are some recommendations inspired by them, and a few taken directly from them.
High talent density should be one of our top priorities. The way we currently think about hiring is: draft a job ad, publish it, advertise it a little, passively wait for the funnel to fill, then run the funnel and aim to pick the best person out of everyone who applied.
Doing it this way, it’s unlikely that we’ll get the best possible candidates for the role, as they probably won’t be applying in the first place. We need to actively look for the best people, and convince them that SaferAI is a great place for them to be. Some pointers to achieve this:
Actively source the best people
- Go through our employees’ networks. Ask them directly, and try tools that make this easier.
- Leverage the talent building ecosystem. We’re lucky that this ecosystem is very rich in AI safety, so let’s use it fully. Ask MATS, Constellation, the Anthropic fellowship, Successif, 80,000 Hours and others for shortlists.
- Ask well-connected people for recommendations. It works much better to ask on a call, or over coffee. In an email, people don’t take the time to pause and think of names.
- When you ask, ask about a specific skill instead of general excellence. “Who is the best risk modeling researcher you know?” is too vague. “Among the risk modeling researchers you’ve worked with, who is best at identifying the most strategic questions and building a research agenda around them?” gets better names.
Build an internal database of exceptional people we’d like to hire at some point
- Over time, build and maintain connections to great people. Say we reached out once for a role and they’re not interested at the moment, but they still like SaferAI. Let’s keep them warm and reach out from time to time.
Hiring should be something every SaferAI employee helps with
- Create a talent Slack channel. Whenever a SaferAI employee comes across someone impressive, whether they met them, read their blog post, or heard them on a podcast, flag that person in the channel. Even if we aren’t looking for that kind of profile now, we can add them to our database and leverage that later.
Think carefully about how to reach out to leads for the first time
- The first reach out is key. If people ignore your first email, they’ll be more likely to ignore the second one. So we need to think carefully about how to contact the lead. What would be most convincing for them to join? (Impact is often a big consideration in the field.) Who should reach out first: someone from the team who has met them, the ED,…? Can we get a warm intro?
- Even if they don’t answer the first message, follow up! This works well in practice. The general idea is to relentlessly pursue good leads, but in a charming way so that you don’t piss them off.
- Proposing a call to tell them more about the role and the org works well. See next point.
Invest in the first conversation with the hiring manager
- Taking the time to talk to a strong lead before they enter the process is worth it. It probably deserves even more time than the first interview. In the podcast, Adam says that one might have to meet a strong lead three or four times before they formally enter a process.
Treat candidates in the pipeline well
- Generally treat candidates in the pipeline with care (don’t leave them waiting too long for an answer, take time to answer all their questions, be extremely clear on the process,…).
- If we go after the best people in the field, the people we interact with are the ones whose opinion of SaferAI gets passed around and sticks. Everyone we talk to, including the people we reject, should come away having had a great experience with us.
Snowball your way up
- This one is recommended by Sam Altman’s post. Ask someone to name the best person they know, then ask that person the same question, and so on.
- I was originally skeptical of this technique, but it has worked well. I now write in my email pitches: “Even if you are not interested, I’d still love to pick your brain on other potential fits.” And that works. I’ve had a lot of calls with people who don’t want to change jobs right now but spend time giving me more names.
Something to flag is that this takes time, you’ll need to deliberately set aside capacity. One number to anchor you: Altman writes that founders should spend between a third and a half of their time on hiring.
Thanks to Alix Pham and Nada Hegazy for their input and help on this draft’s post.