Saffron

Engineers now use AI daily to ship code, but most technical interviews either ban AI outright or can't measure how candidates actually use it. Companies are left choosing between take-homes and live pairing, which don't scale and burn out interviewers, or algorithmic puzzles that reveal nothing about real building ability. Interviewers see only the final output, not whether a candidate architected the solution, pasted it from ChatGPT, or iterated thoughtfully. The result is a score without understanding, testing for a version of the job that no longer exists.

Saffron lets companies connect a GitHub repo or use pre-built templates, define a custom evaluation rubric, and send candidates a single link. Candidates work in a browser-based IDE with native Claude Code access on the company's actual codebase, building and debugging real features with no local setup. Every keystroke and AI interaction is captured, and 10+ independent AI review agents score submissions against the custom rubric. The platform classifies every line of code as human-written, AI-generated, or AI-modified, offers full session replay, and includes AI-checked debrief questions to verify understanding, aiming to replace the entire interview loop with zero interviewer hours and results in hours.

AI changed how engineers ship code daily, but interview processes still test for a job that no longer exists, either banning AI or failing to measure how it's used.