I Write Less Code Than I Used To. That May Be the Point.
Over the last year, my day-to-day job has changed in a way I am still trying to understand. I am still an engineer. I still design systems, read code, debug failures, review implementations, and sometimes build things myself. But I write much less code than I used to, and that feels weird. For most of my career, producing software was the visible evidence that I was doing my job. You had a…
In the last year, the author's job as an engineer has evolved significantly. Once an engineer, the author now designs systems, reads code, debugs failures, reviews implementations, and, surprisingly, writes less code. This shift feels peculiar as, for years, writing code was the tangible evidence of their job. The direct relationship between effort and output is no longer the norm.
A large portion of the implementation work is now being generated with the help of AI. While AI can produce a functional implementation, the real challenge lies in determining whether the feature behaves correctly under various conditions. The expensive part involves ensuring the feature works reliably across the range of actual use cases.
The author spends more time contemplating potential failures, such as unexpected inputs, component disagreements, wrong evidence, biased evaluators, and silent provider changes. They spend time thinking about what "good enough to ship" truly means. Testing and validation, previously seen as downstream activities, are now becoming more critical.
With AI, correctness is more complex and requires engineering. Traditional software offers a reliable contract for deterministic outputs, but AI generates plausible but potentially incorrect outputs. The author's work now revolves around creating reliable AI systems by detecting failure modes, building evaluators, adding policies, and teaching agents to identify errors.
The boundary between creating intelligence and ensuring reliable probabilistic intelligence is becoming increasingly blurred. The author's value is now defined by their ability to define constraints, architecture, failure taxonomies, release criteria, and deterministic guards around probabilistic systems. The frontier of the author's work is no longer simply creating code but defining unknown problems and engineering solutions for them.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
