A procurement brief for an AI-ready design system

Ask three providers for “an AI-ready design system” and you can get three proposals with the same headline and different work inside. One delivers a Figma library. One delivers design tokens and an MCP server. One delivers all of that plus the checks that stop drift. The prices differ because the scopes do, and without a brief nobody can tell which is which.

A brief names every part, so each provider has to say what is in and what is out.

Copy it into a document, replace every bracket, and delete the lines that do not apply. For a first call, a small team can stop after “What goes wrong”; the rest can wait for the proposal.

# Brief: an AI-ready design system for [company]

## Where we are
- Product and surfaces: [web app / marketing site / mobile / email]
- Design system today: [none / Figma library only / code library / both]
- Design tokens today: [none / CSS or Sass variables / a data file, format: ]
- Components today: [count, framework, typed or not]
- Agents and tools in use: [Claude Code / Cursor / Copilot / v0 / other]
- Who fixes agent output now, and hours in a normal week: [role, hours]

## What goes wrong
- Example 1: [link] What was wrong: [ ]
- Example 2: [link] What was wrong: [ ]
- Example 3: [link] What was wrong: [ ]

## What we expect to receive
- [ ] Design tokens as data, in a named format
- [ ] Components with typed props, in [framework]
- [ ] Agent access: [MCP server / static files], hosted by [us / you]
- [ ] Docs agents can read: an llms.txt and a Markdown copy of each page
- [ ] Written guidance: an AGENTS.md or rules files for [tools]
- [ ] Checks that fail output outside the system, run by [CI / the agent]
- [ ] A test set of prompts, with scores before and after
- [ ] Docs and a runbook for whoever maintains it

## Limits
- Stack and repositories: [ ]
- Licenses we cannot accept: [ ]
- Data that cannot leave the company: [ ]
- Deadline, and what it is tied to: [ ]
- Budget range, if set: [ ]

## How we will decide
- Accepted when: [the checks and the score it must reach]
- First phase: [one surface, fixed length]
- Who decides: [name]

What goes wrong is the most useful part. Pick pieces a person had to fix, and say what was wrong: a color typed by hand, a button style that does not exist, a layout that ignores your grid. A provider who reads them can tell you whether the fix is a prompt, a rule or a whole system.

Hours in a normal week should be measured, not remembered. Ask the person who fixes the output to log one week. It is the figure the result is judged against later.

What we expect to receive is a list to tick, not a list to take whole. A team whose agents all run in one editor may not need a hosted server; static files it can fetch may do. The AI-ready design system checklist shows which parts you already have.

Licenses we cannot accept saves a rebuild. If your legal team rules out a font license or an icon set, say so before anyone builds on it; licenses and usage rights lists where the limits usually are.

• Who does the work, from the first call to the handoff?
• What do we own at the end, and in what form?
• How will you show the agent’s output improved: which prompts, which checks?
• What happens to the system when the engagement ends?

With Simbolik, the first answer is Simon Martin himself, from the call to the handoff; the rest come in a one-page proposal after a 20-minute call, with the job, the standard the work has to meet, the price and the timeline, agreed before work starts. Send the brief ahead of the call, or bring it. Work with Simbolik (opens in a new tab).