HR and hiring

How we hired a CTO in a week: a hiring process built on agents

Aleksandr Lukashevich August 25, 2026

I needed a CTO. Not "let's look around for a couple of months", but fill the role fast, because the product couldn't wait. In the end the whole cycle took a week: three days to gather candidates, two days of conversations, two to decide.

Below is how it worked, step by step. This isn't about a magic button. Almost all the manual work stayed manual, it just stopped being mine.

Where the time really goes

When people say "hiring takes forever", they usually mean the search. In practice, time goes into three things, and the search isn't the hungriest of them.

The first is messaging. Reach out to thirty people, get a CV from each, agree on a time, remind, reschedule. Each message takes a minute, but there are hundreds of them.

The second is preparing for the conversation. Read the CV, figure out who you're talking to, come up with questions for that specific person. Fifteen minutes per candidate, and that's fast.

The third is memory. After the tenth call of the day, candidates blur together. Who talked about on-prem, and who talked about metrics? A day later you won't remember.

All three are mechanics. That's what you should hand to an agent, not the choice of a person.

Step 1. Where candidates came from

Two channels worked: a LinkedIn post and going through my own network. I simply went through people I know and asked for recommendations. In three days I had more than thirty responses for the CTO role.

There's almost no automation here, and that's fine. For senior roles the channel matters more than the funnel: strong people come through recommendations, not job ads. The agent comes in where volume begins.

Step 2. Messaging and scheduling

Thirty people had to be handled in a couple of days. I connected an agent to my Telegram and gave it one task: collect CVs from everyone and schedule a half-hour screening call over the next two days.

The agent handled the messaging on its own: said hello, asked for a CV, offered slots, recorded what was agreed. The two days ended up packed tight, twelve hours of calls back to back.

Step 3. A profile before the call

Next comes the part that made the whole thing work. I set up a separate chat called "Interviews" and asked the agent to drop a card for each candidate there, in the order of the calls. Not as a list at the end of the day, but as each slot came up.

The card had four things:

• Grade: A, B or C, marked "borderline" if the data was contradictory.
• Short summary: three or four lines on who this is, what they're strong at, where the risk is.
• A link to the person and the call time.
• A file with questions for 30 minutes, split into time blocks.

The last one turned out to matter most. The questions weren't generic, they were for that specific person. If a candidate had lots of short projects, a question about what they had taken to scale. If they had strong infrastructure and a weak product side, a question about product decisions.

Step 4. The conversations

I recorded each call as a voice note right in the same chat and attached a screenshot from the call. I remember faces better than names. Thanks to the ready-made structure, I never once went over half an hour.

Twelve hours of conversations back to back is a bad idea, and I don't recommend repeating it. By evening the quality of the questions drops, and the quality of the decision with it. If I did it again, I'd spread it over three days of six hours each.

Step 5. Screening at the end, not the start

Usually a recruiter screens out half at the CV stage, to save time. I deliberately moved screening to the very end, and that turned out to be the key decision of the whole process.

After all the calls, the agent transcribed the voice notes and put together two summaries: a table of all candidates with scores by criteria, and a detailed card for each one, with strengths, risks, quotes and dealbreakers.

And here's what came up. Among those a standard process would have screened out by CV were some of the strongest: people with real product pipelines who just can't write a CV. And the other way around: nicely designed CVs with nothing behind them in the conversation.

Did I realize this during the calls? Probably yes. Would I have remembered two days later? Definitely not.

The criteria in the table were weighted: speed, readiness to take on the product, experience with AI agents in production, motivation, teamwork, stack, availability. Some had double weight, because they matter more to us than the rest. The score didn't choose for me. It lined candidates up for a second conversation.

Step 6. Second round and offer

Out of thirty, five were left. A second call with each, going deeper, and the same cycle: profile, structure, recording, summary. The person I chose in the end is still working with us.

A side catch

On top of the summaries, the agent put together two files I hadn't thought about at first.

The first: what candidates said about Mary. Thirty technical people looked at the product with fresh eyes and named its weak spots. Free research that companies usually pay for separately.

The second: a review of me as an interviewer. Where I asked leading questions, where I talked more than the candidate, where I didn't push far enough. Unpleasant to read, useful to apply.

What carries over to volume hiring

The CTO story is a one-off. But three quarters of the mechanics carry over to entry-level roles almost unchanged, and there the volume is many times bigger.

• First-round messaging and slots. With fifty applications a week, this frees up several hours for a recruiter.
• A candidate card before the call. CV + job description + questions for the role.
• Transcript and summary afterwards. Especially if the decision is made by someone other than the person who did the interview.
• One table with criteria. It also ends the eternal argument of "I felt he was stronger".

Separately, something people rarely count: candidates stop waiting. Strong people get snapped up within days, and a reply a week later means you've already lost. Speed of reply here isn't about convenience, it's about access to people.

What not to hand to an agent

The decision. A score is a queue, not a verdict. A person hires a person.

Judging someone by their CV. That's exactly what we moved away from by putting screening at the end. Automatic rejection on a formal criterion is the most expensive kind of saving: you'll never know who you lost.

The conversation. Voice screening is still a person's job, because half the signal isn't in the words.

Personal data without order. CVs, call recordings and transcripts are data about real people. They need a place to be stored, a retention period and someone responsible. Having an agent doesn't cancel that, it makes it more pressing: there's more data, and it shows up faster.

Where to start

The order I'd repeat:

1. Take the messaging off your plate. Collecting CVs and agreeing on times is the first thing to go, and it goes entirely.
2. Make a candidate card before the call. Even without scores, a summary and five questions for the person already change the conversation.
3. Record conversations and transcribe them. Without this, step five is impossible.
4. Build one table with criteria and weights. The decision is still yours, but you'll have everyone in front of you, not just the last three.
5. Screen at the end. If the volume allows, talk to everyone.

And before you start, measure three numbers: how many applications a role gets, how much time passes from application to first reply, and how long the role stays open. A month later the same three numbers will show whether anything changed.

See what can come off your hiring workload

We'll connect to your messaging and count how much time goes into scheduling, preparation and summaries.

Discuss an audit