AI Product Studio
Studio Getting Started
3 pages · scroll straight through
What the AI Product Studio Is
The AI Product Studio is a governed environment where AI agents do the execution and your team sets the intent, reviews the output and owns every decision that matters. It is a peer product to the Product Workspace — not a higher tier of it.
Workspace or Studio?
Both ship product. The difference is who is in the driving seat.
| Product Workspace | AI Product Studio | |
|---|---|---|
| Who executes | Your team | A governed agent fleet |
| You start with | A backlog | An intent |
| Humans spend time on | Planning and building | Setting intent and reviewing |
| How AI is used | Optional bolt-ons assist on request | Native — agents run the delivery loop |
| Method | Agile / SAFe | AI Delivery Lifecycle (AIDLC) |
You can run both. An Organisation can hold Workspaces and Studios side by side, sharing users, branding, billing and audit.
The delivery loop
Studio work moves through a repeating loop. Every step is a page in the Studio sidebar:
Beyond the loop sit the Hybrid Team (humans and agents together), Context & Governance (policies, DPIA, risk notes, audit) and Insights (AI effectiveness and cost).
Where to go next
Creating Your First Studio
A Studio is a self-contained delivery environment with its own members, agents, context, policies and audit trail. Creating one takes a few minutes; the decisions below are the ones worth pausing on.
Create the Studio
From Manage Studios, choose Create Studio and give it a name and a one-line purpose. If your account belongs to an Organisation you can attach the Studio to it now, so it inherits branding, users and billing.
Pick a trust level
Trust Settings decides how much an agent may do before a human intervenes — from proposing a plan only, through to running verified changes end to end. Start conservative; you can raise it once you have seen the agents work.
Configure a model provider
Model Config is where you connect an AI provider. Keys are encrypted with a dedicated AWS KMS key, stored server-side and never reach the browser — test the connection from this page rather than pasting a key anywhere else.
Seed the Context Library
Agents ground their work in the Context Library. Add your product brief, architecture notes, coding standards and anything else an agent should treat as fact. Thin context is the most common cause of disappointing runs.
Assemble the hybrid team
Add people under Humans and choose agents from the Agent Roster. Agent Recruitment walks you through picking agent types for the work you expect.
Set policies and checklists
Policies and DoR & DoD define what a work package must satisfy before it starts and before it is accepted. These are enforced at the gates, so they are worth getting right early.
Intent to Shipped — the Happy Path
The shortest complete journey through the Studio: one idea, one work package, one agent run, one review, one release. Everything else in the Studio is a variation on this loop.
Capture the intent
Open the Intent Pad and describe what you want in plain language. An intent is a want, not a specification — the Studio's job is to turn it into one.
Shape it
On the intent's page, write the brief and let the Composer break it into workstreams and Work Packages, grounded in your Context Library. Review the breakdownsition here — correcting it now is far cheaper than correcting the output later.
Promote to a work package
Shaped work becomes a Work Package — the unit agents execute against. It carries its own checklist, acceptance criteria and lifecycle. Work packages move across the board as they progress.
Run it, and clear the plan gate
Dispatch the package from the Run Board. The agent produces a plan first and stops. A human approves the plan before any work happens — this is the single most important control in the Studio.
Review the output
The package’s Definition of Done reads the pull request: its agents say what falls short and send it back to the coding agent. The Review feed shows everything they are waiting on — rework, questions to answer, stale evidence — and Verify collates what they found across the studio, next to your own CI.
Release and learn
The Release plan puts completed packages on a release, approved on two keys, and names the key results it should move; Release comms drafts what each audience needs to know. Afterwards, Outcome Telemetry and the Outcome Feedback Loop judge whether the release actually moved what it said it would.
When the happy path does not happen
- The plan looks wrong — reject at the plan gate and improve the context, not the prompt. Most bad plans are a context problem.
- Output fails review repeatedly — tighten the work package's DoR and DoD so the gate catches the problem earlier.
- Runs cost more than expected — Capabilities and Model Config hold the guardrails; AI Effectiveness shows where the spend is going.