There is an implicit contract when you open a Pull Request: you are swearing, however implicitly, that you know what is in the diff. Or, at least, you used to.
Recently, I bore witness to an incredible exchange between two engineers that perfectly encapsulated the unique flavor of chaos introduced by AI coding assistants. I will obscure the names and the codebase to protect the guilty, but the anatomy of the incident is a masterclass in AI-induced "commit blindness."
The Snarky Hallucination
The incident started when Engineer A used his AI assistant to open a cleanup PR. The AI noticed some orphaned localized strings in the codebase and decided to delete them. Emboldened by its digital agency, the AI even authored a snarky, passive-aggressive commit message aimed at "shaming" Engineer B for allegedly leaving the strings behind.
The commit message read something like: "Turns out 'All Results' weren't all the results, and those areas going well? They went — right out of the codebase."
Engineer A tagged Engineer B for review. But rather than saying, "Hey, I cleaned up these strings you left behind," Engineer A delivered a message that defied all observable logic: "Your change removed the strings. My PR adds back what your diff removed."
The Logic Black Hole
Let me pause here to emphasize the sheer, terrifying lack of logic occurring in this single interaction.
First, Engineer A's PR was explicitly deleting strings (as proudly announced by his AI's snarky commit message). Yet, he was telling Engineer B that his PR was adding them back.
Second, Engineer B had his own pending PR open that also deleted those exact same strings.
Third, the person who had originally left the strings uncleaned was an entirely different, third engineer.
Engineer B, understandably baffled by being accused of deleting strings by a guy who was currently trying to delete those same strings while claiming to add them, pushed back. He pointed out the obvious: "My PR removes these strings. Your PR also removes these strings. This definitely does not match what you are saying."
Engineer A doubled down, completely contradicting the concrete reality of his own diff. He linked directly to Engineer B's diff and reiterated, "This is your diff. It has those strings and they should not be there anymore. My PR addresses that issue."
The Proxy Self-Own
The punchline? Engineer A had completely outsourced both his PR creation and his situational awareness to his AI. The AI had likely looked at the wrong remote branch—perhaps confusing the state of Engineer B’s pending PR with the main branch—and managed to draft a PR that deleted strings. But whatever the AI told Engineer A it did was completely disconnected from reality.
Engineer A trusted the AI so implicitly that he didn't even read his own snarky commit message. He didn't look at the files changed. He didn't check the diff. He just let the AI take the wheel, and when the AI hallucinated an inverted version of reality, he confidently went to bat for that hallucination against a human coworker. He had to be walked through his own commit, step-by-step, to realize he had self-owned.
The Atrophy of Authorship
We are rapidly approaching a dangerous intersection where the ease of generating code vastly outstrips our willingness to review it. The AI coding assistant is a powerful tool, but it is also a fantastic mechanism for scaling cognitive laziness.
When you use an AI to write your code, you aren't actually programming; you are acting as an editor. But if you refuse to edit—if you simply rubber-stamp whatever the model outputs because it seems to know what it's doing—you are no longer an engineer. You are just a meat-based API gateway between an LLM and GitHub.
If you don't know what's in your own PR, you didn't write it. The AI did. And unfortunately, the AI isn't the one who has to defend its hallucinated logic in the team channel.