# What data do I need before building an AI agent?

3 kinds. The live records the agent reads, like stock, prices or open invoices. The rules it must follow, written down. And a few dozen real past cases to test it on. You rarely need a data project first. You need access to the systems your team already uses, and someone who knows the rules.

By Surinder Thakur, AI Product Designer and Builder. Published 17 Nov 2026. Topic: [Building an AI tool](https://www.surinder.design/work?type=building-an-ai-tool).

## Do I need to clean my data first?

Rarely all of it. The agent only needs the fields its job touches. Clean those, and leave the rest.

A Blueprint finds the gaps early. Missing data shows up in week 1, when it is cheap to fix.

## Why do written rules matter so much?

Because the agent cannot read your senior team's heads. Limits, exceptions and approvals live there today.

Writing them down is often the most valuable part of the project. Your team gets a clearer process, agent or not.

## Why past cases?

Because they are the test. Run the agent on cases your team already decided, and compare its answers with theirs.

In the Ordering agent design, it acts alone only after matching sales reps on 90% of past orders.

## What if a key piece of data does not exist?

Then the agent asks a person for that step, or you buy the data. The ordering design licenses part fit data, for example.

Either way, you find out in the first 2 weeks, not halfway through the build.

## In 1 picture

An agent needs records to read and rules to follow, both before the build.

Data it reads:
- Live records, like stock
- Access to today's systems
- A few dozen past cases

Rules it follows:
- Limits, written down
- Exceptions, named
- Who decides the rest

## Next questions

- [Can an AI agent work with our old systems?](https://www.surinder.design/work/can-ai-agents-work-with-old-systems): Usually yes, without replacing them. An agent connects through what the system already offers.
- [Which task in my business should get an AI agent first?](https://www.surinder.design/work/which-task-should-get-an-ai-agent-first): Pick a task that repeats every week, is easy to check, and has 1 owner.
- [How do you test an AI agent before customers use it?](https://www.surinder.design/work/how-to-test-an-ai-agent-before-launch): In 3 stages. Replay past cases your team already decided and compare answers.
- [What should an AI agent remember about my customers?](https://www.surinder.design/work/what-should-an-ai-agent-remember): Only what makes the next task better, and what the customer would expect it to keep.

## Questions that lead here

- [Can a small team run an AI agent without an AI engineer?](https://www.surinder.design/work/can-a-small-team-run-an-ai-agent): Yes, if 2 jobs are covered. 1 person on your team owns the rules and decides the exceptions.

Page: https://www.surinder.design/work/what-data-do-you-need-to-build-an-ai-agent
