MVP scope template: one page of invariants before the first prompt
A one page MVP scope template for products built with AI agents: the thesis, one workflow end to end, invariants marked by where they are enforced, and the explicit list of what is not built.
How to use it
- Write it before you touch an agent. If you cannot fill it, the week you would spend building is better spent here.
- Mark every invariant as enforced in the database, the application, or both. Never only the interface.
- Every line under "Not built" has a reason. That reason is what turns the tenth proposal into a one line reply.
# MVP scope: <product>
## 1. Thesis (one sentence)
The user can trust <product> for <what it is for, not what it does>.
Test for every ticket: does this serve the sentence above? If not, it is not built.
## 2. The one workflow, end to end
A path a customer would pay for on its own:
<actor> <does> -> <system does> -> <actor confirms> -> <the record exists, with an audit trail>.
Screens: <list>
Data: <entities touched>
Roles involved: <list>
The second workflow is: <name it here, so it is visibly out>.
## 3. Invariants (hold in every state)
| Invariant | Enforced in |
|---|---|
| <e.g. a pending account reads zero rows of any table> | database |
| <e.g. no model output writes to an entity; a person decides; who is recorded> | database + application |
| <e.g. every business write leaves an audit entry> | database |
| <e.g. sensitive identifiers encrypted before the database, never searchable> | application |
Never "interface only". This table becomes the permissions matrix,
and the permissions matrix becomes the security tests.
## 4. Not built (each with a reason)
- <no search by sensitive identifier>: <reason>
- <no infrastructure for a hypothetical need>: <reason>
- <no queue, no cache layer until a measurement justifies it>: <reason>
- <...>
## What this page produces
CLAUDE.md (thesis, stack, hard rules), the domain model with its
invariants, and the permissions matrix. Tickets cite them; they do not restate them.Prompt to use it
Here is my MVP scope (below). Do not suggest features. Instead:
1. Name every place where the workflow implies a second workflow I have not listed under "Not built".
2. For each invariant, describe one state of the product where it would break, and whether "Enforced in" covers it.
3. List product decisions the workflow needs that the page does not take, in the escalation format.
4. Find any line under "Not built" without a reason.
Be terse. One line per finding.
<paste the scope>Why it looks like this, and where each part came from: How to scope an MVP for AI agents.
Changelog
- v1.0Nov 5, 2026First public version. The page took a week to write on a seven week build and was the most valuable week of it.
More from the kit
CLAUDE.md template
The seven sections every agent session inherits, in the order an agent reads them, with nine neutral hard rules to start from.
v1.0 / CLAUDE.mdTemplate / BuildTicket template for agents
Frontmatter a script can validate, plus the three fields that stop agents inventing requirements. Blank and filled versions.
v1.0 / ticket.mdFormat / BuildEscalation and decision format
How an agent stops at a real decision: an id, three options, a recommendation. You answer with a letter.
v1.0 / escalation.mdWorking on a product decision like this?
Discovery, direction and ownership of outcomes is the core of how I lead product work. Explore a product challenge → Or email me.