A rockstar who "failed" our screening

A few years ago, one of my SDE-2s referred someone they called "a rockstar engineer." she couldn't get past our hackerrank test. i was confused, but i respected our process and moved on.

A few weeks later, i saw an update on her linkedin. she'd joined Google.

I went back to my SDE-2 and they practically gave me the "i told you so" face. that moment stuck with me for years.

the candidates you want aren't desperate

Here's what most CTOs and VPs don't internalize: the best candidates are not struggling for callbacks. their work speaks. their testimonials open doors. they have options, and they exercise them.

Yet even these people don't seem to push past the first screening round at many companies. and it's not their fault.

The problem is us.

We're stuck in archaic processes. we built our evaluation pipelines five, ten years ago and never revisited them. we added an ATS, maybe swapped one coding platform for another, and called it innovation.

Meanwhile, the best engineers have moved on. they use AI daily. copilot, cursor, chatgpt — these aren't crutches. they're power tools. and when a company tells them "no AI allowed during our assessment," here's what they actually hear:

• "we don't understand how modern engineering works."

• "we're going to test you on things that don't reflect the job."

• "we value performative coding over real problem-solving."

The best candidates will reject you before you ever get a chance to reject them.

banning AI is fighting the wrong war

A lot of companies right now are spending real engineering dollars on AI-proctoring. they're buying tools that detect whether a candidate used copilot or chatgpt during a coding test.

That's the wrong war to fight.

Think about it. on the job, your engineers use AI constantly. they autocomplete functions, generate boilerplate, ask LLMs to explain unfamiliar codebases, scaffold tests. you hired them to ship outcomes, not to hand-write every semicolon.

So why would you screen for the opposite?

You're filtering for a version of engineering that doesn't exist anymore. and the people who are best at the real version — they see right through it.

what actually matters now

The signal you need from a candidate has shifted. writing code was once a reasonable proxy. it's not anymore.

What matters now is:

• how they direct AI. can they give clear, constrained prompts? do they verify output or blindly paste it?

• how they make decisions. when there are three valid approaches, which one do they pick and why?

• how they explain tradeoffs. can they walk you through what they'd sacrifice and what they'd protect?

• how they debug. hand them broken infrastructure, error logs, and monitoring data. watch what happens.

None of this requires banning AI. all of it requires watching someone actually work.

The companies getting this right aren't asking candidates to implement cycle detection on a whiteboard. they're asking them to figure out why a checkout API is failing for 5% of users — with the codebase, the logs, and every tool they'd normally have.

the real filtering problem

Here's the uncomfortable irony. companies ban AI to "raise the bar." but what they're actually doing is filtering for desperation.

A senior engineer with three competing offers isn't going to spend 90 minutes on a contrived hackerrank problem that tests linked list reversal. they'll just take the offer from the company that respected their time and evaluated them like a professional.

You end up with a pipeline full of candidates who tolerated your process. not the ones who would've been your best hire.

the takeaway for engineering leaders

If your screening process bans AI, you're not protecting quality. you're repelling it.

The developers who are best at using AI — directing it, questioning it, knowing when it's wrong — are the exact people you want on your team. and they're self-selecting out of your pipeline right now because your assessment told them everything they needed to know about how you run engineering.

Update your tooling. evaluate how people think and work, not whether they can perform without the tools they'll use every single day.

Or keep losing your rockstars to google.