We're building assessments for how engineers actually work.

The problem we saw

The three of us were on the job market not long ago. We saw firsthand how easy it was to get through technical interviews without deeply understanding code. Leetcode problems, take-homes, even work trials couldn't accurately catch whether a candidate could actually reason about what AI wrote for them. We got through screens we probably shouldn't have. So did a lot of people we knew.

Then we started talking to founders who were hiring. The same story kept coming up: a candidate crushes the OA, passes the tech screen, maybe even does a work trial—and two weeks into the job, it's clear they can't debug anything, can't extend anything, can't work without AI doing the thinking for them. Interviews were testing the wrong thing, and every round a candidate mistakenly survived cost the company real engineer-hours.

The breaking point was when we realized this isn't just a “nice to have” fix. In late 2025, AI tools were released that have changed the coding workflow so fundamentally that the entire hiring process is testing for skills that no longer reflect the job. A good engineer can now leverage coding agents to be magnitudes more efficient. But a great engineer is one who knows when to trust AI, and when to override it. Unfortunately no assessment on the market measures that.

What we built

Saffron gives companies an end-to-end platform to give candidates a real codebase and the full capabilities of AI tools. We capture everything in a sandbox. Every prompt, every diff, every edit, every decision is scored automatically. Hiring teams get structured metrics on how candidates actually build, not just whether the final output compiles. Candidates get one streamlined platform and an experience that feels authentic to what they'd be doing on the job.