Two AI agents write code, both PRs pass CI, but merging them blows up production. Dive into the Rosentic drama and how it fixes cross-branch conflicts without LLMs.
Delegating your code to AI agents sounds like an absolute dream until they generate 50 PRs that all pass CI but blow up production the second they hit main. Who cries? We do, the poor code-monkeys.
Enter Laramie—a 20-year tech veteran who acts as a tech partnership guy by day and a code junkie using Claude Code and Codex by night on the exact same repo.
He slammed into a massive wall:
Why? Because Git merges text, not logic. And CI evaluates each branch in complete isolation. Nobody—and I mean nobody—was checking if Branch A secretly hated Branch B.
So, he built a tool called Rosentic.
What does it do? It cross-checks every open PR against every other active branch before merging. It catches the semantic breaks that single-PR tools completely miss. The setup is dumb simple: one YAML file, it runs on your own infrastructure, takes 60 seconds to install, and requires zero annoying signups.
The product launched on Product Hunt and immediately got folks talking (currently sitting at 165 points). Here's how the community reacted:
1. The PTSD Tech Leads: A former CTO who used to manage 120 devs jumped in, noting that if human merge conflicts were a massive velocity killer, scaling that up with ai tools working in parallel is a ticking time bomb. Laramie clarified that Rosentic completely ignores textual conflicts (git already handles that) and focuses strictly on semantic breaks. If Branch 1 changes a function signature and Branch 2 still tries to call the old one, Rosentic flags the exact file and line before you merge.
2. The Skeptics (Anti-LLM Wrappers): Seeing the buzzword "Deterministic analysis," the hardcore devs immediately demanded to know the secret sauce. Are we talking AST, an SMT solver, or is this just another lazy LLM API wrapper? Laramie firmly set the record straight: ZERO LLMs in the engine. It parses the code structure, extracts symbols, and does exact structural matching. No guessing, no averaging, no API costs. The same repo state gives the exact same result every single time.
3. The Edge-Case Hunters: Another user pointed out a valid flaw: What about false positives when two PRs touch the same surface for totally valid, coordinated reasons? Laramie admitted the tool isn't magic. It currently focuses heavily on structural contracts (signatures, routes, schemas). Minor side effects like logging updates are outside its scope right now. He practically begged users to run it and report back on these edge cases so they can tune the engine.
TL;DR: Rosentic scratches a massive itch that traditional CI/CD pipelines completely ignore: cross-branch logic compatibility.
What's the lesson for us devs? Never blindly trust green CI checkmarks, especially when you start letting AI scale your output. A green CI doesn't mean your app won't crash upon merge.
Generating code with AI is crazy fast, but managing the semantic conflicts between multiple AI agents is what will actually eat up your mental RAM moving forward. Adapt or get left behind, folks.
Source: Product Hunt (Rosentic)
