The pipeline was never the problem

For two years I said the same thing every hiring manager says: we would love a more diverse team, the pipeline just is not there. Then a recruiter I trusted showed me our funnel. The pipeline was fine. Candidates from underrepresented groups were entering at roughly the rate you would expect. They were leaving at the technical screen, at nearly twice the rate of everyone else.

The problem was not who applied. It was what we did once they did.

Where the funnel actually leaks

I went through the interviews. The technical screen was a live coding exercise, forty five minutes, an interviewer watching. Candidates who had done a lot of these did fine. Candidates who had not, which correlated with who had been given fewer shots at companies like ours, froze. We were measuring interview practice and calling it ability.

We replaced it with a take home that mirrored real work, capped at two hours, followed by a conversation about the choices. Pass rates converged within a quarter. Nothing about the candidates changed.

The other leak was the debrief. Interviewers used words like "culture fit" and "not sure he would gel," and nobody asked what those meant. I banned the phrases and required every concern to be tied to a specific answer to a specific question. The debriefs got shorter and the decisions got better for everyone.

Structure is fairness

Every change that improved diversity on my teams also improved hiring quality overall, and the reason is that they were all the same change: more structure, less vibe. The same questions for every candidate. A written rubric before the loop starts. Interviewers who score independently before they talk to each other. A hiring manager who reads the scores before reading the names.

None of this is exotic. It is the difference between an interview and a conversation that ends in a feeling.

Inclusion is what happens after the offer

Hiring gets the attention. Inclusion is quieter and matters more, because a diverse team that people leave is a sourcing program with a leak at the end.

The things I watch: who gets interrupted in design reviews, and by whom. Who gets the visible project and who gets the maintenance work. Whose idea it was, according to the meeting notes, versus who actually said it first. Who is on the pager rotation and whether it is fair. Who speaks in the team meeting and who only speaks in the 1:1.

I have caught myself on every one of those. The pattern I was worst at was routing the interesting work to the people who asked for it loudly, which reliably meant the same three people. The fix was to stop waiting for people to ask, which is also what the growth conversation is for.

What I say when someone pushes back

Someone on the team, usually a strong engineer who has never thought about this, will eventually say that we should just hire the best person. I agree with them, and then I ask how we know who that is. The answer is the process. If the process measures interview practice, or comfort with a particular interviewer, or whether someone reminds us of ourselves, then "the best person" is the output of a broken instrument.

That framing has ended more of those arguments than any statistic.

The AI rollout, and who got left behind

When we moved to an AI first way of working, I watched who adopted first. It was not random. The early adopters were the people with the most slack: senior, established, confident that experimenting on company time would be read as initiative. The people who waited were the ones who had learned to be careful, and several of them were the same people I had worked hardest to hire.

So the rollout plan changed. Office hours became explicitly a place to be bad at it. The first cohort of peer demos included people who had struggled, showing what they had learned. And I tracked adoption by team and, privately, by who was and was not showing up in the analytics, so that "everyone is using it" did not quietly mean "the usual people are using it."

Inclusion during a change is not a separate program. It is noticing who the change is easy for, and deciding that is not good enough.