Navian Open Source · engineering
Blog.
Field notes on making software reliable in the AI-code era — deterministic testing, reliability tooling, and the occasional one-character bug that cost us five days.
August 25, 2026 · 4 min read
1,000× more testing. Not 1,000× more tokens.
If AI writes 10× the code, you don't need 10× more tests — or 1,000× more tokens generating them. You need one invariant a machine explores across 1,000× runs.
Read →August 17, 2026 · 3 min read
10× productivity by AI coding agents needs 1,000× the testing.
AI writes 10× the code — fast, and it looks right. What it can't do is sense the failures that only surface when timing and bad luck collide. Why the bottleneck moves from writing code to proving it holds up — and how deterministic simulation testing scales for nothing.
Read →