Owning the Verdict, Not Writing Code

The engineer's scarce skill is no longer coding speed—it’s judgment backed by evidence. Addy Osmani frames a shift from model capabilities to loop engineering and software factories, where humans set intent, inspect evidence, and accept risk. Generation scales faster than comprehension, making answerability the bottleneck.

Sonar’s survey shows AI-assisted code is now mainstream, yet 96% of engineers distrust it and only half always verify. This distrust without bandwidth creates a verification crisis. Clean code helps agents too, but review capacity hasn't kept pace with generation—the danger is code no one can explain shipping to production.

Osmani warns of three failure modes: cognitive debt (eroding understanding), cognitive surrender (blind acceptance), and orchestration tax (managing many agents consumes attention). The fix isn't fewer agents but intentional attention design—where humans enter the loop, what evidence they require, and how decisions get owned.

Accountability is the moat with the longest half-life. Skills decay with each model release, but credibility and the signature on shipped work compound. For leaders, the simple rule—explain it or don't ship it—shifts the bottleneck from “can we build it” to “should it exist, and who owns the result?”