Building a Distributed SQL Database in 30 Days with AI

The headline sounds like a flex, and it partly is — building a distributed SQL database is a legitimately hard problem. But the more interesting thing in Kelly Sutton's writeup isn't the database itself. It's what the process reveals about where AI-assisted engineering actually helps.

Where AI assistance changes the math

Distributed systems have a lot of well-understood components: consensus protocols, write-ahead logs, query planners. The concepts are documented; the implementations are not simple. This is exactly where AI assistance has high leverage.

If you understand the concept but would otherwise spend a week translating it into working code, having a model that can close that gap changes your velocity substantially. You're not offloading the hard thinking — you still have to understand what you're building and why. But the implementation time compresses.

Sutton's writeup is honest about this. The AI didn't design the database. It helped him move faster through the parts that were execution, not design.

The 30-day constraint

The time constraint is doing useful work here. A month is long enough to build something real and short enough that you can't afford to get lost in rabbit holes. It forces prioritization in a way that open-ended projects don't.

I'm less interested in the exact stack he used than in the constraint-driven approach. "Build this hard thing in N days with AI as a partner" is a useful experiment structure. The results tell you something concrete about where the leverage actually is.

Worth reading if you're thinking about how to run similar experiments.