System prompt and user prompt
The user prompt is what the person in the conversation writes, and it changes with every message.
The system prompt is written by the company, is the same in every conversation and takes priority. When the two clash, the model should follow the system prompt, and checking that it does is part of the job.
Why it matters
It is the cheapest thing to adjust and what changes an agent’s behavior most.
Before considering training a model or switching provider, there is nearly always room here. Most strange answers come from ambiguous instructions, not from the model.
What it has to carry
Five blocks.
The last one is what separates a careful agent from a dangerous one.
- What the product is and who it serves.
- What information it may use to answer.
- What tone and length it answers in.
- What it must never do, such as inventing figures or promising deadlines.
- What to do when a fact is missing, which is to say so and hand the conversation to a person.
Say what to do, not only what not to do
"Do not invent prices" works worse than "if you are asked a price and it is not in the context, say you will confirm it and hand the conversation to a person".
A prohibition leaves the model with no way out; an alternative gives it one.
It is code, even though it looks like text
It changes behavior in production, so it deserves version control, review and a way to check that a change does not break what was working.
Editing it by hand in a panel with no history is like deploying without a repository.
