The interview for the AI era

See how they actually work.

Coding interviews test whether someone can recall algorithms from memory. That's not the job anymore. Trace puts candidates through real work with a real AI assistant — and captures the reasoning trail of how they think, delegate, and decide.

Not “can you code without help?” — but “how well do you work with AI?”


The problem

The old interview is testing for a skill that no longer matters.

LeetCode grinds and whiteboard algorithms measure memorized recall under artificial constraints. Engineers don't work that way anymore — they work alongside AI, every day. And a green/red pass/fail is opaque to the non-technical founders who most need to hire well.

The old way

Whiteboard & LeetCode

  • Tests memorized algorithms, isolated from real tools
  • Bans the exact AI tools engineers now use hourly
  • Reveals nothing about planning, judgment, or taste
  • A pass/fail diff a non-technical founder can't read
With Trace

Real work, traced

  • Real features, a real codebase, a real AI assistant
  • Measures how they direct, review, and steer AI
  • Surfaces decomposition, judgment, and velocity
  • A readable trace anyone on the hiring team can follow

How it works

Three phases of real work.

Candidates move through the actual arc of engineering — plan, review, build — each one instrumented to surface signal a pass/fail test can't.

01

Planning

The candidate scopes and plans a real feature — with an AI assistant and web search at their side. No blank whiteboard; the real messy start of any project.

RevealsProblem decomposition, judgment, communication, and how they use AI to think — not just to type.
02

PR Review

The candidate reviews a realistic pull request against a real codebase — the kind of code they'll spend most of their week reading, whether a human or an agent wrote it.

RevealsCode judgment, the instinct to spot issues, and how they reason about code they didn't write.
03

Build

The candidate builds live in a real cloud IDE alongside an AI coding agent — fast and smart models, with subagent orchestration. Actual shipping conditions.

RevealsHow they direct AI, when they delegate vs. do it themselves, how they steer and review agent output, and real velocity.

The signal it captures

It's called Trace for a reason.

Behind every phase, Trace captures the reasoning trail — the telemetry and the transcript of how a candidate thinks and orchestrates. The output isn't a score. It's a picture of how someone actually works.

How they direct AI

The prompts, the corrections, the re-scoping. Whether they lead the agent or get led by it.

When they delegate

The judgment call on what to hand off and what to own — including how they wield subagents.

How they review

Whether they catch the subtle bug in generated code — or wave it through. Taste, made visible.

Real velocity

How much working, considered progress they make under real conditions — not puzzle speed.

Judgment & taste

The decisions that don't show up in a diff — scope cuts, tradeoffs, knowing what "good" is.

Readable by anyone

A transcript and trace a non-technical founder can actually follow and reason about.


Who it's for

For the people making the hire.

Hiring managers

Move past trivia and see the real thing: how a candidate plans, reviews, and ships alongside AI — the way your team actually works.

Founders

Hire your first engineers with confidence, even under pressure. Trace turns the interview into a clear, comparable read on how each candidate operates.

Non-technical founders

You can't read a diff — you don't have to. Trace gives you a readable trace of how a candidate thinks, so you can evaluate engineers without being one.

Request early access.

Tell us a bit about you and we'll be in touch — and you can stop testing for a skill the job no longer needs.