The hiring question I left blank

In January 2019 Apres needed someone to lead machine learning. I wrote a short doc called “Hiring Protocol.” It asks three questions. I answered two.

The two I answered

The first question was why this role, and why now. My answer:

The Lead ML/AI Engineer/Co-Founder role is a gap we should aim to fill on the team. We need someone on that side who we can build a team around.

“Hiring Protocol,” January 2019

I wanted deep research or academic experience (a PhD, a master’s, or the equivalent in practice), or deep technical experience. Someone comfortable pushing code at first, who wanted to move into management and build a team.

The third question asked what problems to screen for, if you think of everyone as a mix of skills, values, priorities and personality. I wrote that they should be honest and hard working, driven more by building products that matter to people than by money. Product-focused, self-driven, management experience, and the patience to grow a small company large over five to ten years.

All of that still reads right to me. None of it was checkable.

The one I didn’t

The middle question was this:

What would we like this person to have done in the classic 30-60-90 days for us to feel happy that this was a good candidate?

“Hiring Protocol,” January 2019

There is nothing under it. The doc goes straight to the next heading.

That blank is the most important line in the doc, because it is the only question that would have told us, after the fact, whether we chose well. “Honest” and “product-focused” describe a person. A 90-day outcome describes work you can look at. Without it, you judge a hire the way you judge anything you didn’t define in advance: by how it feels.

Meanwhile, the pitch

The rest of my hiring material was about selling the job. My call script from November 2018 opened with “I appreciate you taking the time to chat for a few minutes. Is this still a good time?” and had a slot for the specific thing in each person’s profile that stood out to me. My recruiting emails for the co-founder roles listed our traction in one sentence: a few hundred signups for the beta, our labor supply, our first enterprise customer.

There is a lot in my notes about how to describe Apres to a candidate. There is nothing about how either of us would know, three months in, that it was working.

What happened

By the time we announced our seed round in January 2021, the team included a Chief AI Officer, a two-time published author and researcher. The plan I wrote for 2020 hired in dated blocks: two sales development reps between January 15 and February 15, one client success hire between February 15 and March 15, then two more sales reps and a developer between March 15 and April 15. Dates for when to hire. Still nothing written down about what each hire should have done by day 90.

What I would write in the blank now

I write every step I hand to an agent with two lines at the end. “Done when” is something a stranger could check. “Evidence” says exactly what to keep. It works for people too. For that 2019 role, it might have looked like this:

Day 30   Measure our labeling accuracy on a held-out set
         Done when: anyone on the team can rerun the number from one script.
         Evidence: the script, and the number it printed.

Day 60   Ship one model that pre-labels a customer batch
         Done when: its labels on that batch are checked by people and the error rate is recorded.
         Evidence: the batch, the checks, the rate.

Day 90   Hire or line up the first person on the ML team
         Done when: they have a start date.
         Evidence: the signed offer.

The specifics would change with the person. The habit wouldn’t: decide what done looks like before the work starts, and write it where both of you can see it. An agent will satisfy exactly what you write down. People are more generous than that, but they also deserve to know what they’ll be judged on.