Tunnel Vision A developer reports that Claude and Codex, while generally producing mergeable code, repeatedly exhibit three failure modes: "prop drilling" data through every layer of the stack, writing unit tests that only verify trivial operations like string concatenation, and over-disclosing in comments and user-facing text — including a support document that repeated "not implemented yet" next to every unfinished button and setting. The developer attributes the pattern to tunnel vision, where the task in hand is the only thing in context, and describes a partial fix: running a separate agent that never saw the original task on a schedule to maintain the support site as a whole from a bundle of recent changes. The developer disputes Cognition's argument that splitting work across contexts loses implicit decisions, contending that decisions living only in context mean the code and supporting materials aren't self-evident. I opened a support document the other day for a feature that was still being built and found "not implemented yet" scattered through it. Not once, as a note at the top, but next to every button and setting the agent hadn't got to. It was accurate, but obviously it had missed the point. Claude and Codex are both moving up the developer curve, and for the most part they write code I'm happy to merge. But there are a few things they both do that I keep hitting:. 1 "Prop drilling". The agent needs a piece of data at location X, so it threads that data through every layer between here and there, clobbering interfaces on the way. The case that stuck with me was expanding our error logging to return specific codes rather than a generic error. The agent drilled every fact the codes needed down through the stack, when at the point the error left the system all of those facts were already sitting there. 2 Unit tests for 1+1. The frontend ended up with a pile of tsx tests checking that string concatenation works. They're cumbersome, they cost build minutes