July 27, 2026
108 PRs and one giant comment war
108 PRs in eight days: Accidentally discovering loop engineering
One coder let an AI churn out 108 code changes, and the internet instantly split
TLDR: A developer says an AI-assisted workflow helped ship 108 code changes in eight days on a small, high-trust team. Commenters were split between impressed and horrified, with the big fight being whether this is a breakthrough in speed or just a faster way to create future messes.
A software developer says an automated helper pushed 108 pull requests—basically bundles of code changes—in just eight days, a wild jump from their usual 5 to 10 a week. The setup sounds like sci-fi office labor: the system grabs a task, writes the code in a separate sandbox, opens a request to merge it, checks that automated tests pass, handles feedback, and lines it up for the human to test later. The author argues this only works because the team is tiny, talks constantly, and doesn’t rely on formal code review as the main way people stay in sync.
But the comments? Absolutely not calm. One camp basically said, “Congrats on discovering how to mass-produce slop,” with critics warning that speed means nothing if nobody is carefully reading the code before it lands. Another commenter sneered that bragging about raw PR counts is like bragging about line count: technically a number, emotionally meaningless. The most dramatic fear was future cleanup—people picturing exhausted engineers wading through piles of overly cautious AI-written code once the honeymoon ends and subscription prices explode.
There was some amused support too: a wry “Power to you friend!” paired with a parental warning to slow down and refactor before the mess comes due. And in peak internet fashion, one user didn’t argue at all—they just dropped a backup link because the original site was timing out. Even the website seemed overwhelmed by the discourse.
Key Points
- •The author says an agent-driven workflow processed 108 tasks in about eight days, versus a previous average of 5 to 10 pull requests per week.
- •The workflow depends on a small, high-trust team where mandatory human pull request review is not required, while the author still reviews changes through testing and outcomes.
- •The author says batching releases through CI and a preview or staging environment works better for this setup than continuous deployment on every merge.
- •The article frames the process as "loop engineering," with a markdown-based protocol and separate orchestration loop and code-writing worker.
- •The system evolved from a single markdown task board to one markdown note per task, using Obsidian's Bases and frontmatter to reduce editing collisions and manage work across multiple repositories.