Don't thank the LLM: the tool's credit erodes your responsibility
Sh0ny
2 августа 20262 min read
In short
Colleagues are increasingly pointing out that a text or test was written by an LLM. It sounds like honesty, but in reality, it’s a way to shift blame in advance. I’ll explain why sharing credit with a tool is a bad idea for an engineer.
Something strange has started happening in our teamwork: when a colleague is asked how many tickets have come in for a new feature, he replies, “According to the LLM—17.” When he publishes a pull request, he adds, “The LLM wrote the unit tests.” It’s as if a writer were to include the version of the spell-checker with every text, or a pilot were to announce the version of the autopilot software over the intercom.
The first impulse is to chalk it up to novelty. But when the habit persists, it’s worth asking: why announce the use of an LLM at all?
The author of the original text lists several reasons. Some believe that credit should be given where it’s due. Others feel like impostors because the result is too good. Some simply didn’t proofread the conclusion and shift the burden of finding errors onto the reader. And some report on the LLM because that’s what they were told to do from above—and “I used AI” grants them a higher degree of leeway for carelessness.
But the main point is about the first case—about “giving credit where credit is due.” Sharing credit with an LLM means sharing responsibility as well. You can’t hold a tool responsible for the result, which means the credit doesn’t belong to it either. When you say “the LLM wrote this,” you’re not being modest—you’re diluting your own accountability in advance. If the result is excellent, take full credit. If it’s sloppy, take full responsibility and improve it.
For engineers, this has a very practical implication. A reviewer who sees “the LLM wrote the tests” subconsciously lowers the bar for review—since it’s a machine, it’s forgivable. An author who cites the tool in advance gets an alibi for code they haven’t read. This is how PRs are born where no one knows which 20% is missing or exactly where it is.
In the pre-LLM world, no one shared responsibility with a compiler or linter. A tool is an extension of your hand, not a co-author. If you’ve decided to release the result to the world, it’s yours. Period.