今天找工作的成本

This one is short, but it’s been bothering me for a while.

A few months ago I stepped up as Head of Engineering, and with the role came the part of the job I’ve always found the hardest: hiring. I’m lucky to already work with great people. But in a startup there’s a constant, quiet pressure to keep looking for talent, because the moment you actually need someone, you can’t afford to wait three months for the perfect candidate.

Skills aren’t the whole story

I used to have grand ideas about meritocracy. People should be hired on their skills and nothing else. I couldn’t have been more wrong. How you behave, how you relate to others, how you approach problems and the energy you bring to a team matter as much as your technical ability, and often more.

From my admittedly anecdotal and narrow point of view, once you clear a baseline of competence, your willingness to do the work takes you much further than knowing a bit more. In a startup this makes sense. It’s how small teams with limited capacity manage to take on giants with thousands of employees, companies that can outbid you for every engineer on the market.

The blockers that have nothing to do with you

That’s why it hurts when good engineers, the kind that are genuinely rare, can’t join us for reasons that have nothing to do with them as engineers. They can’t get a visa. They live in another city. The contract terms don’t work for them.

Lately, though, something else has been on my mind. I believe using LLMs is now an essential part of software engineering. So in our interviews and technical challenges, we expect candidates to use them.

Now put yourself on the other side

Imagine you’re the one looking for a job. You have no income. Job hunting today is largely a numbers game, so you might be interviewing with five companies in a single week. And every one of them hands you a challenge where you’re expected to use AI.

AI isn’t free. The tools that are actually good at coding need a monthly subscription or API credits. So you’re paying for them at what might be one of the worst financial moments of your life.

The usual response is: “It’s always been like this. You always needed money to get a job.” And that’s true. You need housing, clothes and food just to be in a position to apply. But none of those exist only to get you hired. LLM costs are different. They’re ephemeral, and their only purpose is getting the job.

There’s no clean answer

This puts me in an uncomfortable spot. I can accept the status quo, knowing we’ll lose a non-zero number of good engineers simply because they couldn’t afford a decent model for their challenge. Or I can try something different.

I’ve thought about this a lot, and there’s no satisfying solution. Giving candidates a company account for the challenge is off the table: it’s a liability, and I can’t guarantee how it will be used. Reimbursing what they spend doesn’t work either. It’s hard to justify as a business expense, and it quickly drifts into a financial gray area.

What I can actually change

If I can’t remove the cost, the one thing I do control is the process. So the goal becomes making candidates pay as little as possible, and only when it really matters.

Move the challenge later. Nobody should burn tokens for a company that hasn’t had a proper conversation with them yet. If you reach our challenge stage, it’s because we’re already seriously interested, and the money you spend is far more likely to turn into an offer.

Make it smaller. A challenge that needs a top-tier model running for hours isn’t testing the candidate, it’s testing their wallet. I’d rather give a tight, well-scoped problem that works fine on a free tier or a basic plan, then spend the rest of the time talking about how you approached it. That conversation is where I learn the most anyway.

Do it together. This is the one I like most. Instead of a take-home, a live session where the candidate drives and the AI runs on our side, on our machine and our account. No liability, nothing to reimburse, no gray areas. And I get to see what I actually care about: how you think, how you communicate, and how you react when the model confidently hands you something wrong.

Weigh the code less. Honestly, the biggest change is how much the challenge counts at all. These days I give far more weight to a proper conversation about architecture and how you like to build software: how you’d structure a system, where you’d draw its boundaries, how you think about testing, reviews and shipping. When an LLM can write most of the code, the code tells me less about you. How you reason about trade-offs tells me a lot more, and that conversation costs you nothing.

None of these are perfect. Live sessions are stressful for some people, and a smaller challenge shows less. But I’d rather choose a trade-off on purpose than pretend the problem isn’t there.

If you’re hiring too, ask yourself how much your process costs someone who has no income right now. And how many good people you might be losing without ever knowing it.

• * *

Small footnote. If you live in Berlin and you are comfortably coming to our office in Rosenthaler Platz and you love startup environments, shoot me an email to gerard (at) konvoai (dot) com. As a conversation starter, tell me what you think about RAG, how you've used it in the past and if there might be something better for agentic AI assistants. :)