A CTO whose company just raised asked me for 20 minutes on hiring. We went two hours. Most of what I told him I learned the expensive way — scaling an engineering organization to 150 people and paying for nearly every mistake on the list at least once. Writing it down here so the next conversation can start further along.
None of this is theory. It is the set of rules I would hand a founder about to go from 20 people to 100, roughly in the order the pain arrives.
Read one book on hiring#
Who is the one hiring book worth your evening. It covers about 80% of what you need: define the outcome before you define the role, score candidates against a written scorecard, run the same structured interview every time, and chase references harder than you chase candidates. If you read nothing else on the subject, read that.
The rest of this post is what the book does not tell you, or what I only actually believed after it cost me something.
Hiring does not get easier as you grow#
Hiring is the one function that does not get easier as you scale — not until much later than you would expect. Every level you add introduces a role you have never hired for, evaluated by a manager who has never hired for it either. Volume goes up, the criteria get less familiar, and your own minutes per candidate go down. Somewhere past 100 people the machine starts to behave like a machine. Before that it is founders and a lot of duct tape, and pretending otherwise is how you end up with a quarter of bad hires.
Referrals were our highest-ROI channel by far#
Referrals are the highest-return sourcing channel we ever ran, and it is not close. We paid aggressively for referrals that stuck, and most of our best long-term hires came through it. Aggressively means an amount that makes the referrer stop and actually think about their network instead of forwarding a job link — and paid on retention, not on signature.
But "who do you know?" is the wrong question. Ask it and you get five names: the five people anyone would name, three of whom are unavailable and two of whom you already know.
Run a network download instead#
So we ran what we called a network download. Sit with a new hire a few months in, open their LinkedIn connections, and go through it page by page, screening together. Names and photos unlock 10x more than memory does. People remember someone they never worked with directly but heard was great at one specific thing — that is the person you want, and they will never surface from an open-ended question.
It is time intensive. A download runs an hour or two per person, and you do it with every new hire while their network is still fresh. Worth every hour.
Mine the interviews themselves#
Interviews are a sourcing channel, not only a filter. Candidates drop names constantly — the staff engineer who unblocked their migration, the manager who ran the best planning process they have seen. Whenever someone sounded exceptional, we wrote the name down and went after them, including when the candidate in front of us did not pass.
That is how I found Sergio Oliveira Bergmann, my cofounder. We were interviewing my other cofounder at Mesmer, Lucas Silva, who kept bringing this guy up. I texted our head of people mid-interview to say we should hire that other guy. We did.
Firing fast is a fantasy#
Firing fast is a fantasy for almost every founder who repeats it. Firing is brutally expensive, financially and emotionally, and most founders cannot pull the trigger anyway, so bad hires linger for quarters while the team quietly routes around them. It is much easier to be a company that hires slow than a company that fires fast. Build for the first one and treat the second as damage control.
Build a real interview process#
An interview process is a written artifact, not a feeling. So many founders refuse to build one. Everyone thinks they are good at interviews; almost nobody is. They ask about your weekend and decide whether they would grab a beer with you. That is not a process. That is vibes.
This has been settled research for decades. Schmidt and Hunter's meta-analysis of selection methods puts the average predictive validity of a structured interview at 0.51 against 0.38 for an unstructured one. The gap is the structure, not the interviewer. Three things, minimum:
- Define the interview criteria before you start, in writing.
- Prepare the questions you ask every candidate, and actually ask them in the same order.
- Make every interviewer write down what they learned, so you can compare candidates instead of comparing impressions.
Never hire by consensus#
Consensus hiring is the practice of giving everyone on the panel a veto, and it assumes something that is almost never true: that every person interviewing understands what you are looking for and where you want the culture to go. They do not. One person owns the decision, including the power to override a no.
I learned that the hard way. A candidate got glowing reviews from the whole team. I saw red flags in 40 minutes. I deferred to the group because they had spent more time with him. I fired him a month later, and every single thing that went wrong was something I had seen in that interview. I never ignored my gut again.
Urgency from a hiring manager is usually the wrong kind#
Urgency is a founder's favourite drug, which is exactly why it is so easy to sell one. A manager tells you the team desperately needs a role filled and you should start hiring now. You ask what they are looking for and they say there is no time to write the job description, let's just get candidates into the pipeline.
Founders are suckers for urgency. Injecting it is part of the job, so when someone matches that energy it reads as progress. It isn't. The right urgency is knowing exactly what you are looking for, having someone who can actually evaluate this function, and knowing how the process runs. Writing that down takes under an hour with an LLM. One interview with an unqualified candidate costs you more. Build the muscle to push back.
Trust the gut, then move fast#
Your gut is an early-warning system with a terrible interface. If it says someone isn't working out, you are almost certainly right. Most founders are analytical and want proof first. By the time you have proof, the damage has been compounding for months.
This gets worse as you scale. If a whisper reaches you, that person has been underperforming for a long time before anyone felt safe enough to say it out loud. Get as close to the situation as you can — read the work, sit in the meeting, talk to the people next to it — and then decide fast.
Generous severance is what makes speed possible#
Generous severance is the policy that turns a decision you have already made into a decision you can actually execute. Make 3 to 4 months of severance a company policy. For most VC-backed companies that is not a big number, and it buys you and your managers a clear conscience: you stop agonizing over whether it is fair and start acting. Netflix built most of its culture memo around the same trade — part ways quickly, pay for it properly.
And when you let someone go, actually let them go. No staying three more weeks to hand things off. People disengage the moment they realize they will not have to live with their own decisions. Rip the bandaid off, let the team grieve, move on.
Skip levels are the highest-signal meeting you can run#
Skip-level meetings are the best instrument I ever had for knowing what was actually happening in the org. As you get further from the day to day, you need your own pulse on the team beyond what managers tell you. Quarterly skip levels were the highest-signal meeting I ever ran.
One question does most of the work: who on your team is doing such a great job that you would do anything to keep working with them? Ask ICs that and you will learn more about performance and team health than any reporting chain will ever tell you — including the names your managers have not mentioned.
The same instinct applies to the numbers. Measure the org and the systems around your team rather than grading individuals, which is the argument Lucas makes in life after DORA. And if you are hiring engineers right now, the profile that wins has moved: range, judgment, and fluency orchestrating agents, which is what the fastest teams were selecting for in Q1.
The short version#
The short version is the one I gave him at the end of the two hours. Hire slow on purpose, because you will not fire fast. Write the criteria down before you meet anyone. Put one owner on every decision. Pay real money for referrals and go get the names yourself instead of waiting for them. Make severance generous enough that acting on a bad hire costs you nothing but money. And build a channel into the org that does not run through your managers, because by the time bad news reaches you through the chain, it is old.
If this was useful, the rest of these notes live on the Field Notes column — what we are hearing in CTO conversations, written down as it happens.