beat claude or don't apply

The four things I look at when someone tells me they're AI native.

August 14, 2026

Hiring in 2026 has become an absolute mess.

Everybody writes AI native on a resume now, so it’s hard to determine if someone is really “AI native”.

The solution?

I ask candidates to open their harness on the interview call. Cursor, Claude Code, Codex, whatever they run. I want a few minutes of watching them work inside it.

Four things come out of those few minutes.

  • Usage levels. Someone who lives in it every day is maxing out their usage limits. That number is hard to fake.

  • What's in production. Anyone can build anything right now, so a folder of demos tells me nothing. I want the thing that's running and being USED by someone other than them.

  • GitHub activity. Ask them to open it. An active account is hard to fake.

  • Which model they run. When someone is still on an old model, I ask why. If the answer is "the new one came out too recently," that's a reasoning problem. How you do anything is how you do everything.

Then we run the Beat Claude challenge, which we forked from Anthropic's version.

Applicants have to beat the model at a job specific task. It's a fast read on how someone thinks.

We're hiring right now. If you want to come work with me, take the challenge here. :)


One founder I had dinner with recently went further and cut his entire recruiting team.

What replaced it is a personalized landing page for each person he sources. Filled with the projects you'd work on, who you'd work with, and a few questions if you want in.

His reasoning: the people worth sourcing are good enough that they'll never jump through a normal hiring funnel, so he removed the funnel and did the research himself.

Genius right?

Try it on your next interview call. Ask them to share their screen and see how fast you know if they’re really “AI native” or not.

To the people who show their work,

Eric Siu