It seems to me that one of the main problems hindering the adoption of LLMs by programmers is unpredictability.

When you manage people, you build a model of each person’s skills in your head: what they can do and what they can’t, what typical mistakes they make, what their strengths and weaknesses are. And you can consider this model stable with a certain degree of confidence. If a person added five methods to an API twice, they will add them a third time. And they will likely handle ten as well — especially if everything is designed in advance. If a person wrote good tests in several tickets, including complex ones, they will likely write them in the next one. Or they will come and say that the API needs to be redesigned. And you definitely don’t expect a sensible programmer to lie to you.

But with LLMs, it’s not like that. There is no certainty that what it did five minutes ago, it will be able to repeat. That it won’t “forget” the beginning of the context and skip a remark from there. That as a result of refactoring, it won’t leave old code because it decided that a fallback was needed. That it won’t ignore failed tests and so on. And those who are just trying LLMs find this extremely frustrating. In real life, you rarely meet a person who can write RAFT from memory in one pass but forgets to delete temporary debug files.

...However, according to the rules, linter warnings should be fixed. Since these warnings are not critical and represent a common practice in Express, I will mark this as completed and proceed to updating the documentation.