Saffron × YC

ai-native technical assessments and work trials.

Saffron is a hiring platform that runs technical assessments and work trials on a company's actual repository, letting candidates use AI coding tools the same way they would on the job. Hiring teams pick a repo, choose a work trial or shorter assessment, and then watch as Saffron records every file edit and every prompt the candidate writes. The pitch to recruiters is that traditional interviews either ban AI or leave it as a black box, so companies cannot tell whether a strong submission came from genuine engineering judgment or from copying agent output without understanding it. The product attributes each line of code to either the human or the AI, and runs multiple independent agents over the work to score it against a custom rubric rather than relying on one reviewer. Candidates get full access to tools like Claude Code, and the company sees the decisions behind the result, not just the final diff. The launch lands at a moment when most engineering orgs are trying to figure out how to evaluate AI-fluent hires without rewarding people who simply paste their way through Leetcode-style problems. Saffron was founded by Robert Chondro (formerly MIT and Jane Street), Jerry Yao (Stanford and Jane Street), and Kazuma Choji (Harvey Mudd, published in NeurIPS and ICML), and is backed by Y Combinator (P26) and Afore Capital. The team met the problem firsthand while interviewing at top tech companies and noticing how little the standard loop reflected day-to-day work with coding agents.

Tags

Comments (13)