All posts

What I Look For When Hiring a Software Engineer

Code is easy to fake, especially with AI. Passion and problem solving are not. Here's how I vet for them.

I end up having the same conversation a couple of times a year. Someone I know is hiring their first engineer, or they just got burned by a bad hire, and they want to know how I vet people. So I’m finally writing it down.

The short version: I look for two things, passion and problem solving. Everything else is secondary.

Why not just look at the code?

Code is easy to fake, and it’s easier than ever with AI. A polished GitHub profile, a slick take-home project, a portfolio full of impressive-sounding work: any of it can be generated, borrowed, or stitched together from tools the candidate doesn’t actually understand.

Passion and problem solving are much harder to fake. You can’t prompt your way through a real conversation where someone keeps asking “okay, and then what?”

The pitcher analogy

I think about hiring a lot like scouting a pitcher in baseball. Some things you can coach, and some things are just innate ability.

You can’t teach someone to throw 95 mph. Every pitcher has a peak, and if their peak is 85, you could spend a decade coaching them and they will never throw 90. But if they throw 95 and their off-speed pitch is weak, you can teach that. If they’re wild, you can help them find control.

To me, passion and problem solving are the fastball. You either have them or you don’t, and I don’t believe they can be taught. Languages, frameworks, methodologies, tools: all of that can be taught. So I hire for the fastball and coach the rest.

Vetting for passion

This one is pretty simple. I just talk with them about projects they’ve worked on recently.

Passion usually doesn’t show up in what someone does at their day job. It shows up in what they’re building on the side. You can hear the excitement when they talk about something new they’re trying, a tool or service they just found, and what they love about it. Think of a friend who just found a great new restaurant, somehow brings it up in every conversation, and insists everyone has to try it. That’s the energy you’re listening for.

If someone can’t name a single thing they’ve been curious about lately outside of their assigned tickets, that tells you something.

Vetting for problem solving

This is harder, because you usually have to tailor the question to the person’s skillset to really make them think.

I normally start with something like:

“On that project you just told me about, let’s say you were asked to add feature X, something you’ve never done before. How would you go about it?”

Then I just listen to them think out loud.

Find something they haven’t done

The key is to find something they’ve never done. If you ask about something they’ve already solved, they have a recipe stored in their head and you’re testing memory, not problem solving.

If you have their resume and repos, it’s usually easy to spot a gap. A few examples:

  • Infrastructure: If they’ve only ever worked in AWS, never GCP: “Your company just signed an exclusive contract with Google, and you need to move all your infrastructure over to GCP. How would you handle that? What steps would you take?”
  • Architecture: If it’s clear they’ve never worked with queues and workers: “How would you change your system if single requests started taking anywhere from 1 to 20 minutes to complete?”

What a good answer sounds like

I’m not looking for the “right” answer. I want to hear:

  • How they break the problem into smaller pieces, and how they approach each piece.
  • Honest uncertainty. Phrases like “I’m not sure about this part because I’ve never done it, but I’m assuming it would work something like this…” are great signs.
  • Questions back to me. I want them to ask for clarification, because it shows they know how to gather real requirements before they start building.

There are two failure modes on either end. You don’t want someone who makes zero assumptions or guesses and just freezes. And you don’t want someone who acts like they know everything.

Don’t skip the basics

Some of the most revealing questions are the simple ones. Recently we interviewed a candidate who could talk at length about complex AI systems and massive infrastructure. We gave him a simple prompt:

“You’re designing a platform where teachers can create multiple-choice tests and quizzes. Walk me through the database schema. What data would you store, and in which tables?”

He couldn’t answer it.

All he needed to say was something like “a users table with roles, a quizzes table, a questions table, an options table, and an answers table.” That’s it. We would have nodded and moved on, knowing he understands how data should be stored. Big buzzwords are easy. Fundamentals are where you find out what someone actually knows.

The “when do you ask for help?” question

I almost always ask this hypothetical:

“You’re given a task you’ve never worked on before. You do some preliminary research and you’re still stumped. How long before you ask a colleague or manager for help? An hour, a day, or a week?”

  • An hour suggests they may need to be micromanaged.
  • A week suggests they’ll drag things out because they’re afraid to ask for help.

You want someone who puts in a real effort and doesn’t need their hand held, but who also recognizes when they’re wasting time and isn’t too proud, or too insecure, to ask for guidance.

Catching people who overstate their experience

Sometimes a candidate will simply lie, or at least stretch the truth a long way. Problem solving questions are one of the best ways to expose that, because it’s very hard to bluff through specifics.

For example:

“Have you worked with GCP?”

“Yes, I know GCP really well.”

“Okay. You need to set up a simple RESTful API and a frontend SPA. Which specific GCP services would you use, and why?”

Someone who actually knows GCP will answer that easily. Someone who doesn’t will fall apart fast.

A friend of mine hired someone who claimed he’d built “the fastest vector database” for a specific industry, used by several real, well-known companies in that space. Those clients checked out. What my friend found out later was that the whole thing was a drag-and-drop workflow tool wired up to a hosted vector database. The candidate had never built anything outside of visual automation tools.

A tool like that isn’t the problem. Overstating it is. A single follow-up would have exposed it: “How would you have built that if you didn’t have access to those outside tools?” Flip the question, take away the crutch, and see what’s left.

The best interview isn’t an interview

Honestly, almost everyone I’ve been part of hiring, I knew before they ever applied for the job. That’s one of the best parts of meetups, user groups, and community events.

When people aren’t applying for a job, you see the real version of them. An interview is a lot like a blind date: you aren’t meeting the real person, you’re meeting the “best version” they’re presenting. Interviews are basically a performance.

So if you’re going to hire engineers, the best thing you can do is show up in the places engineers spend time when nobody is evaluating them. Then, when you do interview, skip the trivia and the take-home tests. Have a real conversation, listen for the excitement, and give them a problem they’ve never seen before.

The fastball shows up pretty quickly.