Learn · Explained
What is an AI foundation, and why build it first?
Published 31 October 2026
Short answer. An AI foundation is everything your agents need to know about your company: access to your data sources, your business language, what your numbers mean, and how you work and get tasks done, written down so AI can use it. Getting your data in order grounds your agents in your business context and makes them more capable.
Build it first, because every agent you add works from it. A model knows how to reason, but it knows nothing about your company until someone writes that down.
This guide explains the five parts of an AI foundation, how it differs from a data foundation, why it comes before the agents, and how to start.
The five parts of an AI foundation
| Part | What it is | What your agents can do with it |
|---|---|---|
| Access to your data sources | Read access to the systems that hold your figures, through data pipelines and an MCP server. An MCP server is a server an AI tool uses to reach your data, built on the open Model Context Protocol. | Use the data they need, and only that data. |
| Your business language | Your company dictionary: what you call things, what each thing means in your company, and how the things connect. A “customer” is someone with a signed contract; an invoice belongs to a supplier. | Read a question the way your people mean it. |
| What your numbers mean | A map of your data: which system holds which figure, and which one is the official source. | Choose the right data, and know what each number means. |
| How you work | The handbook, which says how things are done here and is read at every run, and skills, which are written instructions for one task each. | Do a task your way, and keep a correction once it is written in. |
| Sources | Every figure can be traced to the record it came from. | Show where every number comes from. |
Each part is a written record your people can read and correct.
The AI foundation and the data foundation
The data foundation is one part of the AI foundation. It covers the data. The rest of the AI foundation covers what your people know about that data.
| The data foundation | The rest of the AI foundation | |
|---|---|---|
| What it holds | Your systems’ data, connected in one warehouse in your own cloud, cleaned, with every figure traceable to its record | Your words, what your numbers mean, and how you work, written down |
| Where it comes from | The systems you already run | Your people’s heads, policies and habits |
| What goes wrong without it | Agents read stale or conflicting figures | Agents read the right figures and still misunderstand them |
How the data foundation is built is in The data foundation, explained.
Why the foundation comes before the agents
Agents answer from it instead of guessing. An agent grounded in your data answers from what it has just looked up in your sources, and shows where it found it. Without a foundation, it has nothing to look up. Read Grounding is the antidote to hallucinations.
You write it once, and every agent shares it. The second agent uses the same dictionary, data and handbook as the first. Setting up the next job becomes easier, because most of what it needs is already written down.
Corrections improve the foundation, not the model. When the operator corrects an agent, the correction goes into the dictionary, the handbook or a skill. The learning stays with your company when you change model.
Much of it pays off without AI. Connected, traceable data gives you reliable reports, whatever you decide about agents. A written dictionary and handbook help a new colleague as much as a new agent.
The foundation and the infrastructure
The AI foundation is what your agents know. The AI infrastructure is the setup that lets them do real work in your systems, safely: where their tools run, the limits they cannot cross, and the person who signs off. Read What is AI infrastructure?
How to start an AI foundation
- Pick one question your company answers every month, for example “What did we spend on freight last quarter?”
- List the systems the answer touches. That list is your first data pipeline. Get your data in order before you buy another AI tool walks through this step.
- Write down the words the question uses, and what each one means in your company. That is the first page of your company dictionary.
- Write down how the answer is worked out today: which figures, which checks, which exceptions. That is the first page of your handbook.
- Give read access to those systems, and nothing more. Your IT, in-house or your IT partner, sets it up; your company decides what is granted and can take it back.
- Check that every figure in the answer traces to its record. If one does not, fix the data before an agent relies on it.
- Add the next question. Each one reuses what the last one wrote down.
What one agent needs on top of the foundation, such as its own records, its tools, its limits and its logbook, is in What an AI agent needs to do real work.
The foundation is yours
The AI foundation sits in your own cloud, and your company controls who and what can read it. You can grant, narrow or withdraw an agent’s access at any time.