# Three levels of AI output

By Tommy Magee

Type: Framework
Canonical: https://tommymagee.com/resources/three-levels-of-ai-output

A shared way to read the complexity of anything built with AI, from a personal tool to an enterprise system. It’s also how a council filters what gets built.

**Use this when:** Teams are building with AI, and you need to judge how complex each output is and how much support it needs.

## Why it exists

Once teams can build their own tools, building stops being the hard part. Other people start relying on what someone made, effort gets duplicated, and nobody is sure who looks after it. Treat every tool the same and you either smother experiments or leave critical work without an owner.

## The framework

1. **Personal tool: Does anyone besides the builder depend on it?** Built for one person’s work. Light guardrails, fast iteration, no formal support. Most of the learning happens here, so it should stay easy.
2. **Shared team solution: Would the team notice if it broke tomorrow?** Other people rely on it. It needs a named owner, a place in the tool registry, and someone checking that it still works and still uses current information.
3. **Enterprise initiative: Does the business depend on it?** It crosses teams or touches critical decisions. It gets engineering support, testing, access controls, and a budget. Moving a tool to this level is an investment decision, not a reward.

## How to use it

1. Keep a tool registry so everyone can see who owns what.
2. Review the registry regularly. Look for tools that have quietly become shared, and for duplicates.
3. Decide on purpose: leave it personal, adopt it as a team solution, invest in it as an enterprise system, or retire it. This is how an AI business council filters the flood of new tools.

## What moves a tool up

- Other people now rely on it, and the owner can no longer tell who they are.
- It touches data that needs access controls or a review.
- It crosses teams, or feeds a decision the business cannot easily reverse.
- Someone with budget agrees it is worth the engineering support. Moving up is an investment decision.

## What retires it

- Usage has dropped and nobody is asking for it back.
- The owner left and no one picked it up.
- A shared or enterprise tool now does the same job.
- The data or model it depends on changed and fixing it costs more than the tool is worth.

## What the registry records

- Name.
- What it does, in one sentence.
- Owner.
- Who relies on it.
- Level: personal, team, or enterprise.
- Data it touches.
- Last reviewed.

## In practice

Our champions built tools for contract review, finance and accounting, meeting records, and dashboards that cut down coordination between groups. As other people came to depend on them, the registry showed which ones needed an owner and more support.

## Where it breaks down

The levels describe how much support a tool gets, not how much it’s worth. A personal tool can matter a lot to one person. And a registry only helps if keeping it current is easier than ignoring it.

## Try it with AI

```
I'm reviewing tools and agents my teams have built. For each one I describe, ask who built it, who uses it, and what happens if it breaks.

Then place it at one level:
- Personal tool: only the builder depends on it.
- Shared team solution: a team relies on it, and it needs a named owner.
- Enterprise initiative: the business depends on it, and it needs engineering support.

Flag tools that people rely on more than their support level suggests, duplicates, and anything that should be retired.
```

Template: /templates/tool-registry.md

Topics: AI champions, AI enablement, AI governance
