VibeHire
Hiring developers in the AI era

See how a candidate actually works with AI.

VibeHire analyzes a developer’s real coding sessions — Cursor, Claude Code, Codex and other agents — and gives you a clear profile of how they build with AI. Read from real work, not a take-home test.

Their code never leaves their machine. Only a privacy-safe summary is shared.
No spam — one email when we open for early access.
How it works

Three steps. No accounts for candidates.

1

You share one link

Create a position and get a single link with a command to run. Send it however you like — email, Slack, a job post.

$curl -fsSL vibehire.in/a/… | bash
2

They run it locally

One readable command reads their real AI-coding history on their own machine.

Code, diffs & transcripts stay put
Only a redacted summary is uploaded
3

You get a clear report

An evidence-based read across five signals — honest about what it can and can’t tell — ending with interview questions grounded in their real work.

Ready to read in minutes
What you get

A report you can read in five seconds

Big, scannable, and honest about its own limits. Bands are a read — never a grade, never a hire/reject score.

A
Afsal
Software Engineer
afsal@vibehire.in
Solid evidenceClear strength
AI CODING ASSESSMENT · 49 SESSIONS · 2,000 COMMITS

Reference-grounded verification shipper

Grounds agents in references, verifies outputs before shipping, and delivers steadily through CLI-heavy, artifact-building workflows across web app and AI chatbot projects.

How much did they ship?
215,029
lines · 2,333 commits · 12 repos
Builder type
Burst
63% short-burst work blocks
Longest streak
21d
consecutive days shipping
Plan-first rate
24%
of 49 sessions
Longest session
12h 25m
22 deep sessions ≥1h
Agents at once
3
peak · 318 subagent runs
Change-course
1%
13 redirects / 367 sessions
Time invested
52h 38m
active · 33-word prompts, 17 a session
Work mix
Fixes
389 features · 528 fixes
Decisions
507
144 held · 36 reversed
How they work with AI
Steering
Steers by anchoring work to concrete references and reusable instructions rather than constant mid-course correction — a 1% redirect rate across 367 traced sessions.
Solid
solid evidence
Verification
One of the clearest signals: repeatedly checks outputs before accepting, asks for before/after proof, and leans on test/build tooling — a validation-first workflow.
Strong
solid evidence
Debugging & recovery
A fix-heavy commit mix (528 fixes), recorded debugging decisions, and command/read/edit loops point to practical, iterative debugging and repair.
Solid
solid evidence
Decision-making
Broad, product-aware decisions across architecture, scope, quality, and debugging — 507 in all; of those with known outcomes, 144 held and 36 were reversed.
Solid
solid evidence
Delivery
Substantial across volume and continuity: 215,029 lines, a 21-day shipping streak, and broad activity across 12 repositories with heavy subagent fan-out.
Strong
solid evidence
See the full sample report →
Privacy by design

The honest part of the deal

Candidates run a readable command they can inspect first. It only reads — it never writes, and it never uploads their work.

Never leaves their machine
  • —Source code
  • —Diffs and commit contents
  • —Full chat transcripts with the agent
  • —Plan files and internal notes

No take-home test, no screen-share, no work sample to write. The signal comes from work that already happened — and the candidate stays in control of what’s shared.

FAQ

Questions, answered

Is this a coding test?
No. It reads work that already happened — real sessions and commits — so there's nothing to schedule and nothing for candidates to cram for.
What does a candidate have to do?
Run one command you send them. It reads their AI-coding history on their own machine, and they approve a short summary before anything is shared.
Do you see their actual code?
No. Their code, diffs, and chat transcripts never leave their computer. You get a redacted, numbers-only summary — never raw code.
What if there isn't enough data?
The report says so plainly and sets that candidate aside. It never guesses, and never scores someone low just for a thin history.
Is it a hire / no-hire score?
No. There's no overall score or ranking. It's there to make the interview sharper — every report ends with questions grounded in their real work.
Which AI tools does it work with?
Claude Code, Cursor, Codex, and other AI coding agents.

Hire for how people really work with AI.

Be one of the first teams to try VibeHire.

No spam — one email when we open for early access.