TECH Signal 477
Bruno Quaresma proposes building features end-to-end before splitting into PRs, citing AI-assisted decomposition
A product engineer argues that AI assistants have made post-hoc branch decomposition cheap enough to replace upfront issue splitting, so teams should build entire features on one branch and slice them into review-sized PRs only after the code reveals natural boundaries.
The workflow inverts the traditional RFC-to-issue-to-PR pipeline by deferring decomposition until after implementation, trading pre-planned structure for earlier product validation through working demos. The judgment half of code review and product validation remain bottlenecks that AI does not solve, which is why narrow PRs are still necessary. This is a single author's perspective carried by one feed, not a corroborated industry shift.
Written by elseif from the cluster below · every claim links back to a sourceThe three things worth knowing
AI assistants made building, design interrogation, and splitting finished branches into sequential PRs dramatically cheaper, removing the main reason to pre-decide boundaries before coding.
The workflow keeps design first, running an adversarial interrogation skill called grill-me and committing specs before code, but builds wide on one branch with save-point commits that are discarded before review.
Narrow PRs remain necessary because the judgment half of review, including subtle correctness and architectural fit, has not gotten cheaper, and reading narrow diffs is how engineers own code they did not type.
THE CLUSTER