InfluxDB co-founder Paul Dix ran AI coding agents on real production tasks, watched what failed, pulled back to writing code by hand, and then returned to AI-assisted development with heavier oversight. That full arc, from optimism to retreat to cautious re-engagement, is the actual story here.
The interview details which projects made it to production and which did not, and why the failure modes matter more than the wins. Dix also wrote 'Build the machine that builds the machine' and posted on X about '2026: the great engineering divergence,' two threads that frame his argument about where the skill gap between developers is actually forming right now.
This is worth reading in full for the specifics: what Dix sent the agents to do, what broke, and what his current oversight process looks like. The conclusions are practical, not philosophical.
[READ ORIGINAL →]