The Reflex to Defer to LLMs Is a Documentation Problem

"Just ask an LLM" has become the default deflection when someone asks a technical question — in Discord servers, GitHub issues, and internal Slacks. The post argues this is not a productivity win; it is a documentation failure wearing a helpful face.

The core problem: LLMs are generative, not authoritative. They produce plausible answers, not correct ones. For questions that touch edge cases, version-specific behavior, or undocumented internals, the model has no reliable ground truth to draw from. Sending engineers there first does not save time — it introduces a verification step that would not exist if the documentation were adequate.

For solo founders and small teams, the damage compounds. When institutional knowledge lives only in human brains and model weights, it is not really documented. It is just deferred. The next engineer who hits the same wall goes through the same lossy retrieval cycle.

The argument is not anti-LLM. It is anti-laziness dressed as tooling adoption. LLMs are useful for exploration, scaffolding, and rubber-duck debugging. They are poor substitutes for a well-maintained changelog, a precise error reference, or a decision log that explains why the architecture looks the way it does.

The practical implication for engineering teams: if your first instinct when someone asks a question is to tell them to use a chatbot, that question is a documentation ticket. Log it. Write the answer down somewhere findable. The LLM can help you draft it.

For developers building AI-adjacent products, this is also a product signal. Users who distrust LLM answers on consequential questions are not wrong to distrust them. Retrieval-augmented systems that cite primary sources outperform raw generation precisely because they address this gap — not by being smarter, but by being auditable.

Good documentation is not a legacy practice that LLMs obsolete. It is the input quality that determines whether LLM-assisted workflows produce reliable output at all.

Source