Prompt engineering
Prompt engineering is the work of writing the instructions that tell an AI what job to do, what format to use, what it must not do, and when a person should take over.
Prompt engineering is the work of writing the instructions an AI system follows for a job. The prompt sets the task, the format, the sources it may use and the conditions that should stop it. In a chat window that is a question you type once. In a business process it's a saved brief that runs on every similar case. You test it against real examples and keep it with the process, not in someone's notes.
In operations the prompt sits behind a repeating job. A classification prompt tells the system how to sort incoming requests and where to send anything it cannot place. An agent prompt lists the steps and the systems it may use. It also names the actions that need a person. You run both on a sample of real cases before they go live. Staff should not rewrite the live brief each time they use the tool. When the process changes, the prompt has to change with it.
Prompts fail when they are vague. Telling the system to be helpful does not say what to do with a missing field or a request outside its remit. They also fail when a prompt written at launch sits unchanged while the live process has moved on. A tighter prompt cannot invent a policy you never supplied. It also cannot reach a file it was never given.
When it matters
- →The same AI job produces different answers from different staff.
- →An AI agent is about to run a defined operational process.
- →Helpdesk replies need a consistent format and a clear stop point.
- →Launch prompts have not been updated since the process changed.
Related terms
Prompt engineering: common questions
What is prompt engineering?
Prompt engineering is the job of writing and maintaining the brief an AI follows for a piece of work. You specify the task, the output shape, what the system may look at, and what it must refuse or escalate. For a one-off chat that is a typed question. For a live process it's a versioned instruction that every similar case uses. Operations teams should treat it as process documentation, not as a clever sentence someone typed once.
Is prompt engineering still relevant?
Yes. Stronger models still need a brief for the job you want done. They don't know your process, your tone or which fields matter unless you say so. A saved prompt is how you make that repeatable. If every member of staff types a different request, you get different output. The work is to lock the instruction to the process and review it when the process changes.
Does prompt engineering prevent hallucinations?
A clearer prompt can reduce sloppy answers. It can require a source, ban invented figures and tell the system to say when it doesn't know. It cannot create a fact that was never in the files. If the model has no access to the procedure or the record, it may still produce a complete answer anyway. Prompt work sits alongside retrieval from your documents and a person checking anything that updates a record or goes to a customer.
Ready to talk?
Book a free, no-obligation discovery call. We'll learn about your business and show you exactly how Wanzo can help — with a bespoke proposal within 48 hours.
No commitment. No sales pressure. Just honest advice.