raghav@dev:~

When not to use an LLM

AI models are brilliant with messy things like language, tone and fuzzy questions. But some jobs need a simple rule, and using AI there makes things slower, costlier and less reliable.

Here are three signs you need a rule, not a model.

1. The answer must be exactly right, every time. Working out tax on a bill, or checking a password has at least 12 characters. There is one correct answer. A rule gets it right every time. An AI gets it right most of the time, and “most” is not good enough for money.

2. The rule fits in one sentence. “Orders over ₹500 get free delivery.” If you can say it in one line, write it as one line of code. Sending it to an AI is like asking a lawyer to check whether 600 is bigger than 500.

3. You have to explain every decision. If a bank rejects a loan, it must be able to say exactly why. A rule can point to the line that said no. An AI can only give you its best guess at its own reasons.

A simple way to remember it: a calculator for fixed answers, a smart friend for judgement calls. Most good products use both, the AI for the fuzzy parts and plain rules for everything that must never be wrong.

Before reaching for AI, ask: could I write this as a rule? If yes, do that first.