Backed by Y CombinatorGood code is easy now.Good engineers aren’t.
See who's engineering and who's vibecoding. Candidates build on your real codebase with AI, and you see every decision they make.
The Problem
Technical interviews were designed
for a different era.
Engineers ship with AI every day. Most hiring processes still pretend they don’t.
AI changed the job, but not the interview.
Every engineer uses AI to ship — for boilerplate, debugging, architecture decisions. But most assessments either ban AI entirely or can’t measure how it was used. You’re testing for a job that doesn’t exist anymore.
Take-homes don’t scale, and interviewers burn out.
Pairing an engineer with every candidate for hours is the gold standard — but it doesn’t scale. So teams fall back on algorithmic puzzles that tell you nothing about how someone actually builds.
You see the output. You never see the process.
Did they architect the solution or paste it from ChatGPT? Did they iterate thoughtfully or thrash? Today’s assessments give you a score. They don’t give you understanding.
The Platform
See exactly how every candidate builds.
01 — Create Assessment
Connect your repo. Define your rubric. Send a link.
Link your GitHub repo or choose from pre-built templates. Then define custom evaluation criteria with 10+ AI review agents. Invite candidates with a single link.
02 — Candidate Experience
Candidates build on your actual codebase
Candidates get a browser-based IDE with Claude Code on your actual repo — no local setup required. They build, debug, and refactor real code, exactly how they'd work on the job.
03 — Review Results
Know exactly how they built it
Multiple AI agents score every submission against your criteria. See which code was human-written vs AI-generated, replay the full session, and verify understanding through AI-checked debrief questions.
How It Works
One assessment replaces your entire interview loop.
Your current process
With Saffron
0+ agents
Score every submission independently against your criteria.
Every line
Classified as human-written, AI-generated, or AI-modified.
Full replay
Every keystroke, every prompt, every decision — reviewable.
Pricing
Simple plans that scale with your hiring.
Basic
$199
/mo5 assessments / mo
For teams starting to hire with AI.
Premium
Most popular$499
/mo15 assessments / mo
For teams hiring at scale.
Everything in Basic, plus:
Enterprise
Custom
Unlimited assessments
For orgs with high-volume or custom needs.
Everything in Premium, plus:
Need more assessments? $49 per additional assessment.
Includes $5 Claude Code budget per assessment.
Compare
The only platform built for how engineers actually work.
Other platforms added AI copilots. Saffron was built around AI from day one — on your codebase, with full attribution.
What candidates build
AI tools for candidates
AI interaction tracking
How work is scored
Human vs. AI code attribution
Works on your codebase
Interviewer time required
What You Get
What you get from every assessment.
Not just a score. A complete picture of how your candidate engineers.
Multi-agent scoring
10+ independent AI agents evaluate against your custom rubric. No single interviewer's bad day skews the result.
agents per review
Your codebase, their skills
Candidates build on your actual repository. No toy problems, no sandboxes — real complexity, real signal.
AI as a tool, not a crutch
Full Claude Code access, just like the job. Every prompt, suggestion, and decision is captured.
Line-by-line attribution
Every line classified: human-written, AI-generated, or AI-modified.
Full session replay
Every keystroke, every file, every AI interaction. See how they think, not just what they shipped.
Session replay
See how your next hire actually builds.
See every line of code, every AI interaction, every decision — before you make an offer.