AI Product Studio Overview
AI Product Studio is a governed, observable environment in which AI agents collaborate with your team on product work — from intent capture through to release verification. It is a different way of working from the Product Workspace, and the same Free, Business and Enterprise plans cover both.
One plan covers the Studio and the Workspace. Free supports one developer and the default agents; Business adds custom and marketplace agents for as many developers as your Organisation has seats; Enterprise adds organisation-wide agent and integration management. Every plan uses BYOK AI providers.
What the Studio does
Each Studio is a self-contained AI workspace for a product or workstream. Inside a Studio, your team:
- Captures work as intents — free-form problem statements, ideas, bugs, and spikes — then lets AI classify, score, and shape them.
- Runs agents that produce code, tests, documentation, security assessments, accessibility fixes, and compliance artefacts.
- Reviews every change through the work package’s Definition of Done, whose agents read the pull request and whose approval also needs a person; the Review feed shows what each pull request is waiting on.
- Sees verification across the stack on Verify — security, accessibility, performance, compliance, tests — beside each pull request’s own CI.
- Maintains a complete, immutable audit trail linking every released artefact to the intent that created it and the human who approved it.
Why a Studio?
General-purpose AI assistants produce output without context, governance, or accountability. The AI Product Studio provides all three:
| Dimension | General-purpose AI assistant | AI Product Studio |
|---|---|---|
| Context | One-shot prompts, no memory | Grounded in your Context Library, ADRs, and standards |
| Accountability | No approver, no record | Named human approver on every release |
| Traceability | Chat history in a browser tab | Full audit trail from intent to production |
| Quality gates | None enforced | Definition of Ready and Done, merge hold, two-key releases |
| Governance | Ad-hoc | DPIA, policies, model risk assessment |
| Cost visibility | Flat subscription (no per-run insight) | Cost tracked and capped per run |
The Studio Delivery Cycle
- Capture — Intents created on the Intent Pad
- Classify — AI assigns kind, confidence, and priority signal
- Shape — AI decomposes intent into scope, acceptance criteria, and sizing
- Plan — Agent produces a step-by-step execution plan for human review
- Execute — Agent executes with HITL checkpoints on blocking decisions
- Review — The Definition of Done reads the pull request; the Review feed shows what it is waiting on
- Verify — Findings collated by concern and beside each pull request’s CI; the merge is held until the gate passes
- Release — Completed packages released on two keys; comms approved before they go out
- Measure — Outcome Telemetry records the readings each release is judged by
- Learn — The Feedback Loop judges each release against its expected signals and proposes course corrections
Operating modes
- 🧑✈️ Supervised: AI proposes plans; every step requires explicit human approval. Available on every plan
- 🤝 Semi-Autonomous: Agents apply what they write and accept what they suggest, and Auto Pilot is available; runs auto-pass unless flagged, and a person still approves each work package and decides what ships. Business and Enterprise plans
- ⚡ Fully Autonomous: The studio runs the whole loop from discovery to a pull request with no manual input, under the owner's standing approval, given once on Trust Settings and withdrawn when the studio leaves the mode. The loop is started with a spend ceiling on the Autonomous loop page, where everyone in the studio sees what it is doing and steers it rather than drives it: pause, stop, skip an intent, move one to the front. Merging stays with GitHub or a person. Business and Enterprise plans