Working with Intents

Intents are the entry point for all work in the AI Product Studio — natural language problem statements transformed into structured, agent-ready work items.

Capturing Intents

Navigate to Intent Pad and click + Capture Intent. Write your problem in plain language — describe the problem, not the solution. You can also dictate the title by voice, attach a screenshot, or import from a GitHub Issue, GitLab Issue or Jira ticket URL:

✅ "Checkout button is invisible to screen readers — no accessible label"
✅ "Payment API response time spikes over 3s during peak hours"
❌ "Fix the button" (too vague)
❌ "Implement aria-label on checkout button" (solution, not problem)

AI Classification

The classification engine assigns a Kind, Confidence (0–100%), and Priority signal. If confidence is below 70%, add more detail or manually override.

Intents from an Ideas Board

On an Ideas Board, open ✨ Generate, choose Intents and run it. The agent reads the board and proposes up to six intents. Each is a piece of work the studio could take through brief, work packages and pull request on its own once the intents it names are done — with a statement of what exists when it is done, a few signals you could observe, and the board notes it draws on. On a studio with no repository connected, the first is the foundation the others build on.

Implement after — each suggestion says which intents come before it, with a reason. A suggestion with none says so: it can be implemented first.

Context, not intents — technology choices, personas, glossary terms and non-functional requirements are not work to implement, so they are listed separately. Generate context entries from these drafts Context Library entries from the same notes.

Accepting some, or all — tick the suggestions you want. If one you ticked is implemented after one you did not, a dialog names each such link before anything is created and offers to accept the prerequisites too. Carry on without them and the intents are still created; the missing prerequisite is written on the intent’s page so it can be linked later. A link is never dropped silently.

Captured from the board — every accepted intent arrives with a page of that name: what is done when it is done, its acceptance signals, what it is implemented after, and the board notes it came from, quoted as the board held them. It is a proposed page, like any page an agent writes: read it and choose Apply before the agents use it. A Semi-Autonomous or Fully Autonomous studio applies it on arrival.

The order intents are implemented in

An intent’s page has an Implementation order card. Implement first lists the intents this one builds on — add one by choosing it and, if it helps, saying why; a link that came from a board is marked suggested by an agent. Unlocks lists the intents that build on this one; it is worked out from their links, so change it on those intents.

The order is advice. An intent whose earlier intent is not done yet shows a banner, a ⏳ after N chip on its Intent Pad card and a dashed arrow on the map — nothing is disabled or refused, and a lone developer can take any intent forward. The one thing refused is an order that loops. The agents are told the order as fact when they plan or shape the intent.

The Intent Pad’s Map view draws the whole studio: intents left to right by step, an arrow from each to the ones implemented after it — solid once the earlier intent is done, dashed while it is not — and the same order as a numbered list beneath.

From an intent to Work Packages

Open an intent from the Pad to reach its own page. Its short description sits under the name; press ✏️ to edit it in place. Once the intent is in progress its name, kind and description are frozen with the record. Write the brief (Problem / Opportunity and Approach, or have the Architect draft it), set its priority, due date and value, then use the Workstream Explorer: generate workstreams against the brief, and under each one generate or add Work Packages — each a real package the moment you accept it. Open a package to add its work items. A named human must approve every package at its readiness gate before it can be dispatched.

Review acceptance criteria carefully. The AI may miss edge cases requiring domain knowledge. Every criterion must make sense to you.

Intent Status Flow

Captured → Classified → Scoping → In Progress → Done → Archived

Statuses also advance automatically: saving the brief moves Captured → Scoping, and Mark intent complete on the intent page closes it once every work package is coded.

All AI Product Studio documentation

Build v0.3.2 · deployed 2026-10-19T23:52:31Z