A survey of 900+ Pragmatic Engineer subscribers finds codebase quality is declining and most engineering managers do not care. The burden of maintaining increasingly complex codebases is falling on a shrinking pool of engineers who still understand them. Junior engineers are hit hardest: they report AI tools as less helpful while racking up higher token costs, suggesting they need mentorship and room to develop skills, not just autocomplete.
AI adoption at the org level is failing on its own terms. Companies cannot achieve productive team-scale adoption, and survey data shows outcomes depend heavily on the engineering culture that existed before AI arrived. Agenttic tools are producing slot-machine behavior, with pricing structures respondents describe as designed to encourage compulsive prompting. Meanwhile, fewer engineers hold negative views of AI than in 2024, but positive sentiment has not risen by much, even as model quality has improved significantly.
The full article is worth reading for what it documents about code ownership: the concept is eroding, and team collaboration is becoming less important as a result. That structural shift, not productivity metrics, is the real story here. This is Part 2 of a three-part series from The Pragmatic Engineer; Parts 1 and the tooling survey are linked within.
[READ ORIGINAL →]