InfluxDB co-founder Paul Dix ran AI coding agents against real production workloads and reported back with a split verdict: some code shipped, some did not. His core thesis, detailed in his piece 'Build the machine that builds the machine,' is that agentic coding requires a fundamentally different oversight posture, not just a different tool.
The detail worth reading for is Dix's account of why he stopped letting agents write code entirely, then reversed that decision. He returned to manual coding, assessed what broke and why, then reintroduced AI with stricter supervision. That cycle, not the conclusion, is the signal.
His LinkedIn post on 'the great engineering divergence' frames 2026 as a fork in the road for engineers who adapt versus those who do not. The interview sits at the intersection of that prediction and lived experiment. Read it to understand the specific failure modes, not just the outcome.
[READ ORIGINAL →]