Let them use AI, unsupervised
An agent one-shots your take-home in minutes. The final diff looks clean, passes the tests, and tells you nothing — not who actually solved it, not whether they understood a line of it.
Coding interviews for the agentic era
See how candidates work with an agent — not just what they ship.
plaiback gives every candidate a built-in, fully-logged Claude agent inside a real web IDE. Every prompt, tool call, diff, and decision is captured and replayable — so you assess orchestration skill, the thing that actually matters now.
Join wait list See how it works Async take-home (no scheduling) or live — same full record either way.
Replay above is illustrative. Already invited? Sign in →
The problem
Coding tests broke — and "no AI allowed" doesn't fix them.
Algorithmic screens and take-homes were built for a world where a candidate typed every line themselves. That world is gone. Which leaves most teams stuck choosing between two bad options:
Ban AI and lock it down
A take-home ban is an honor system you can't enforce — the strongest candidates quietly use an agent anyway, and you're none the wiser. Lock it down with a proctored screen instead and you're testing whiteboard recall: a skill your own engineers abandoned the day they started shipping with agents.
There's a third option — and it's the whole idea behind plaiback: hand them the agent on purpose, and watch how they drive it. Directing an agent is the job now, and it's the one thing a final diff can never show you.
See how it worksFrom invite to decision in three steps
Take-home or live — evidence-rich either way.
Send an invite
Pick a task from the catalog (or bring your own), set a budget and time window, and send a one-click invite link. No setup for the candidate.
They work with the agent
The candidate solves the task in a real in-browser IDE with a built-in Claude agent and live preview. Everything they and the agent do is logged as it happens.
You replay & score
An LLM judge grades the session against your rubric; you replay the timeline, confirm or adjust, and compare candidates side by side. Decide on evidence.
Hiring signal built for how engineers actually work today
Candidates ship with agents on the job. Your interview should let them — and show you whether they're any good at it.
The event log is the product
Every prompt, tool call, file diff, and agent decision is recorded as a replayable timeline. You watch the reasoning unfold instead of guessing from a final diff.
LLM-judge + human scoring
An LLM judge grades each session against a rubric and surfaces the moments worth a second look. Humans confirm the call — fast where it's clear, careful where it isn't.
Take-home or live
Send it as an async take-home — no calendars, candidates work whenever — or run it live and watch the session unfold in real time. Same complete, replayable record either way.
Real tasks, real sandbox
Candidates work in a genuine web IDE with a live preview and a curated task catalog — trap-based problems where 'green tests' aren't the same as 'correct'.
Replay & compare
Re-run any session step by step, or put two candidates side by side. Calibrate your bar against evidence, not vibes.
Orchestration over typing speed
The skill that matters now is directing an agent well: framing problems, catching its mistakes, knowing when to dig in. plaiback measures exactly that.
Run your next coding interview on plaiback
We're onboarding new teams by hand while we learn what works. Join the wait list and we'll reach out.