Why I Left My Job to Build AssessAI

I had a good job. Full-stack engineer at Avoca, a YC-backed company building AI voice agents. The team was strong, the problems were interesting, the pay was fine. I wasn't running from something.

I was running toward a problem I couldn't stop thinking about.

the moment

It happened during a hiring committee meeting. We were reviewing five candidates for a senior engineer role. Three had failed the coding interview. One couldn't solve a medium-difficulty LeetCode problem. One got the optimal solution but couldn't explain their approach. One panicked and ran out of time.

The two who passed? They wrote working code. Clean solutions. Textbook performance.

We hired one of them. Within three months, he was struggling. Not with coding — with everything around it. He couldn't decompose ambiguous requirements. He'd build what was asked without questioning whether it was the right thing to build. He'd write code but never think about how it would fail in production.

Meanwhile, one of the candidates we rejected — the one who couldn't explain her approach under time pressure — went to another company and shipped a product that handled 50k concurrent users. I know because I saw the architecture post on Twitter.

Our hiring process tested the wrong thing. And I realized this wasn't an Avoca problem. It's an industry problem.

the problem nobody's fixing

It's 2026. AI writes production-quality code. GitHub Copilot, Claude Code, Cursor — these tools are mainstream. The skill that differentiates great engineers isn't syntax recall or algorithm speed. It's product thinking: decomposing ambiguous problems, reasoning about tradeoffs, collaborating with AI tools effectively, and communicating clearly.

No assessment platform tests for this. HackerRank tests algorithms. Codility tests speed. Every platform I researched was still optimizing for a skill that's rapidly depreciating.

I spent two weeks doing competitive analysis. Talked to 15 engineering managers. 12 of them said the same thing in different words: "Our coding tests don't predict job performance, but we don't know what else to use."

That's a market gap. A big one.

the first week

Day 1: Wrote a one-page spec. Five assessment dimensions — problem decomposition, system thinking, AI collaboration, communication, technical depth. The flagship feature would be LLM Interaction Mode: give candidates an AI collaborator and measure how they use it.

Day 2-3: Architecture decisions. Next.js 15, Supabase for the entire backend, OpenAI for evaluation. I wanted a stack where one person could move as fast as a small team. Every service I chose had to handle auth, database, and hosting without me building infrastructure.

Day 4-5: First working prototype. A candidate could sign up, see an assessment, answer questions, and submit. No scoring yet. No AI interaction. Just the skeleton.

Day 6-7: TDD setup. I wrote 40 tests before writing a single feature. This felt slow. It saved me three weeks of debugging later.

what I learned in that first week

The problem is sharper than I thought. Every engineering manager I talked to agreed the current system is broken. But none of them had budget or time to fix it. They want a turnkey solution. That's my product.

Solo speed is real. Without meetings, without PRs, without waiting for reviews, I shipped more in 7 days than I would have in 3 weeks at a company. The tradeoff is nobody catches your mistakes. TDD is the safety net.

The tech stack matters less than the constraints. I could have spent a week evaluating frameworks. Instead, I picked the stack I know best and started building. Next.js + Supabase isn't the "best" stack objectively. It's the best stack for a solo founder who needs to ship fast.

was it worth it?

Eight weeks later, the answer is yes. 275 questions across 20 categories. 630+ tests. A real product that real companies can use to hire engineers based on how they actually work.

I don't know if AssessAI becomes a business. I know the problem is real, the solution is better than what exists, and I'm building faster alone than I ever built on a team. That's enough for now.