# AI is not a system. AI is part of a system.

> I don't learn AI to master AI. I learn it to build and run a business differently.

Author: Aleš Nypl · https://engineloop.cz/ales-nypl

Companies often ask me where they could use ChatGPT. It's a reasonable question, but not the first one. The first question is: what problem are we solving, and what has to change because of it?

## The chain I start from

I always follow the same path: business problem, process, data, technology, AI, automation and measurable impact. AI is only the fifth link. If the first four aren't clear, even the best model won't help.

## A demo is not a production system

Almost anything works in a demo. The difference is what has to hold up every day. Using an app is not the same as calling an API. A prompt is not an agent that has permissions, memory and rules. And a database is not a knowledge base in which only approved facts count. I care about what is under the demo and how it could be run safely.

## How I brief AI

- **Context and problem.** What I'm solving and why.
- **Constraints.** What AI must not assume and what it must verify.
- **Output.** The form I need it in.
- **Verification.** How I'll know it's right.
- **Rules up front.** I write them before a mistake happens, not after.

Rules matter more than the prompt. A prompt is a one-off instruction. Rules always apply and work for me every day.

## Where the person belongs

At every step of the chain I ask what may be automated and where a person has to decide. Where the answer is clear, AI can move fast. Where responsibility is at stake, it must not.

I don't learn AI to master AI. I learn it so that, with its help, I can build and run a business differently.