Vibe AI Explained: Why Letting Candidates Use AI Is the Right Call
Most technical assessments are still built on a premise we think is already wrong: that a fair test means no outside help. We built Vibe AI on the opposite premise, and this is the case for why.
We didn't set out to build a coding assessment product. We set out to hire engineers for our own team, and we kept running into the same problem: the tests we were using to screen candidates measured something our best engineers had stopped doing months earlier. Nobody on our team writes algorithms from a blank editor anymore. They prompt, they review, they iterate, they ship. Our hiring test didn't know that.
The blank editor is testing for a world that's gone
Recent developer surveys put daily AI tool usage among engineers at 87 percent (GitHub Developer Survey, 2024). That's not a fringe behavior to police — it's the baseline. A test that bans AI to keep things "fair" isn't neutral. It's measuring a skill that matters less every year, while staying silent on the skill that matters more: knowing what to ask, when to trust the answer, and when to throw it out.
What we built instead
Vibe AI gives every candidate a built-in AI assistant during the assessment. Nothing hidden, nothing to sneak in. Every prompt they send is logged and timestamped alongside their code, and we score three things separately: whether the code is correct, whether it's well-structured, and whether the reasoning behind it holds up — did they iterate on a weak answer, or copy the first thing that compiled?
That distinction is the entire point. A candidate who asks three precise, escalating questions and lands on a clean solution is doing exactly what we want them doing on the job. A candidate who pastes the prompt and ships whatever comes back, untested, is showing us something too — just not the thing a pass/fail score would have told us.
The honest counterargument, and why we don't buy it
The obvious objection: doesn't this just make it easier to cheat? It would, if we were still scoring like a blank-editor test — final answer, pass or fail. We're not. The prompt log is the assessment. A candidate can't game a reasoning score by getting a lucky output; the questions they asked and the judgment they showed asking them are right there in the report, alongside the code.
Vibe hiring is the idea. Vibe AI is the proof.
We use "vibe hiring" loosely, internally, to describe the broader belief this comes from: that hiring should evaluate how someone actually works, not how well they perform under artificial constraints designed for a different decade. We're not asking anyone to adopt that as a framework on faith. Vibe AI is the concrete, shipped version of it — a real assessment format, in production, generating real prompt logs and real reasoning scores today. Judge the idea by that, not by the name.
Frequently asked questions
Isn't letting candidates use AI during an assessment just making it easier to cheat?
Only if the assessment still scores like a blank-editor test — pass or fail on the final answer. Vibe AI scores the prompt log alongside the code, so copying an answer blindly and iterating toward a real solution look completely different in the report, even if both submissions technically pass.
What is 'vibe hiring,' and is it the same thing as Vibe AI?
Vibe AI is the shipped product: a technical assessment format with a built-in AI assistant and full prompt logging. "Vibe hiring" is the broader idea behind it — evaluating candidates on how they work with the tools they'll actually use on the job, not on their ability to work without them. Vibe AI is the concrete, working version of that idea; it's the only part of this we're asking anyone to take on faith.
Do candidates perform better or worse once you let them use AI?
Scores don't collapse to a ceiling the way a pure syntax test might. The spread we see is in reasoning quality — how a candidate prompts, whether they push back on a bad suggestion, whether they understand what they shipped. That's a harder signal to fake than a correct-looking function.
Why not just require candidates to disclose AI use instead of building a whole assessment format around it?
Self-disclosure relies on a candidate volunteering information against their own interest during an evaluation — it doesn't happen reliably. Logging every prompt automatically removes the need to ask.
Get started
See what a Vibe AI prompt log actually looks like.
Start a $2.99 trial and build your first Vibe AI assessment in under 5 minutes.
Further reading