Why software factories fail, part two
Strong agents and developer-tools item because it turns the AI makes too many PRs complaint into a process diagnosis: the problem is too many bad PRs, not too many PRs.
Logged at IST: 2026-07-26 03:34 IST
What it is: Dex Horthy’s part two of “Why Software Factories Fail,” published as an X Article.
Gist: Dex’s follow-up says the answer to unreliable lights-out software factories is not more faith in model judges. For now, the judge is still the human, so the right move is to put review and planning back into the loop earlier.
He proposes four leverage points: product requirements, system architecture, program design, and vertical slices. Product review pins down the user pain and success condition. Architecture aligns services, endpoints, schemas, queues, and stores. Program design goes one level lower into types, method signatures, call-stack shape, file-tree diffs, and layout decisions. Vertical slices keep the work touchable instead of letting the model build horizontally through thousands of lines before anyone can run or inspect the feature.
The useful diagnosis is that teams may not have too many pull requests. They may have too many bad pull requests. A small amount of upfront planning can make review cheaper than discovering all the important decisions at the end.
Newsletter angle: Strong agents and developer-tools item because it turns the “AI makes too many PRs” complaint into a process diagnosis: the problem is too many bad PRs, not too many PRs.
Retrieval note: Grounded from the FXTwitter API article payload attached to the X post.