AI Product Studio
capabilities in full
Set the intent, let a governed agent fleet execute, and review every output before it ships.
Agents do the execution — people set the intent, review the output and own the decisions.
Delivery Loop 15
The repeating cycle an intent travels: captured, shaped, packaged, run, reviewed, released and measured.
Vision
Set the direction the Studio works towards, so agent output can be judged against something other than the last instruction.
Ideas Board
A live, shared canvas for workshops and retrospectives — notes, groups, connectors, reactions, facilitated votes and templates, with a timer, a lock and optional anonymity. Its structure feeds Generate, which cuts the board into intents that can each be built on their own, says which to implement first, and carries the notes onto each one. Every idea keeps a link to what it became.
Intent Pad
Capture what you want in plain language. An intent is a want, not a specification — turning it into one is the Studio's job. A Map view lays the intents out in the order they build on each other.
Intent shaping
Break an intent into workstreams and work packages against your context, and correct the breakdown before anything is built.
Auto Pilot
In Semi-Autonomous mode, let the Studio run its own generation loops — from one intent through to a reviewable work package per slice — and review the result rather than each step.
Codebase check
Before an intent is planned, read your connected repository for what already exists, so the plan changes what is there instead of building it again and lists what it left out. On an Auto Pilot run a second AI request checks the plan against the ask before anything is written.
Work Packages
The unit agents execute against, with its own checklist, acceptance criteria and lifecycle. Each package draws where it sits: what is implemented before it and what builds on it, across intents. A Delivery stage follows it past dispatch to the coding runs, branches and pull requests that deliver it.
Pull request delivery
Send a package to the Studio's own runner, GitHub Copilot or Claude and follow the pull request. The Definition of Done reads the change, sends shortfalls back to the coding agent, and holds the merge with a commit status until it passes.
Run Board
Every run in the studio with a link back to what started it — filter by what needs you, by agent or by date — and every outstanding pull request with its Definition of Done state and merge hold.
Review feed
Everything the Definition of Done's agents want from each pull request, in one place: rework, questions to answer, stale evidence and merge holds. Answer an agent and the answer can go straight to the coding agent's pull request.
Verify
What verification says across every package in delivery — by concern, by pull request next to your own CI, and finding by finding.
Release plan
Plan releases in work packages whose Definition of Done has passed, grouped by release phase and workstream. Two-key approval, and nothing ships until its pull request has merged.
Release comms
Release comms for customers, support, security and operations, and a formal change record for auditors — drafted from the release record and approved before they go out.
Outcome Telemetry
Record metrics and signals — bugs, regressions, incidents, tickets, feedback — and see which release each counts towards.
Outcome Feedback Loop
Judge each release against the key results it said it would move, at +7 days and at its horizon. A release that falls behind raises a course correction for a person to accept.
Context & Governance 9
The controls that make agent-led delivery defensible: policy, risk, evidence and a complete audit trail.
Code Repositories
Connect repositories so agents can read and propose changes against real code.
Context Library
The material agents ground their work in: product briefs, architecture notes, standards. Thin context is the usual cause of poor runs.
Product Manual
What the existing product already does and how it is built — capabilities, architecture, data, integrations and conventions — derived from your connected repositories with file-path evidence, and updated as work packages are delivered. The Context Library records what you have decided; the manual records what exists.
Recommendations
Agent-raised suggestions that wait for a human decision instead of applying themselves.
Policies
Policy-as-code enforced at the gates, so a rule is a control rather than a note in a document.
DoR & DoD
Work package checklists that decide what may start and what may be accepted. Each item names who reviews it — a person's role, an agent, or both — and Definition of Done items can judge the pull request itself. An item can name the controls it is evidence for (ISO 27001, SOC 2, NIST SSDF, SLSA or your own), and an exception to an item records who granted it, why and until when. Every check and decision lands in the work package's Passport — its evidence record, signed when a release is approved.
DPIA & Model Risk
Data protection impact assessment and model risk records, kept with the work rather than in a separate system.
Risk Note Generator
Produce the risk note for a change from what the run actually did.
Audit Explorer
Reconstruct any run: the intent, the context bundle, the plan, who approved it, what changed and what shipped. Owners and studio admins can also read and export the studio's change log, field by field.
Hybrid Team 5
People and agents on one team, with explicit roles rather than an anonymous pool of automation.
Ask Agent
Put a question to an agent that already has your Studio context, rather than starting from a blank prompt.
Humans
Manage the people in the Studio, their roles and what they are accountable for.
Agent Roster
The agents available to your Studio, what each is for, and the guardrails each runs under. Sync it with your repository: import its Copilot and Claude agent definitions, or export a roster agent as those files through a pull request.
Agent Recruitment
Choose agent types for the work you expect, rather than enabling everything and hoping.
Hybrid Team View
See humans and agents together, so it is clear who — or what — owns each part of the work.
Learn 1
Whether the agents are actually earning their place.
AI Effectiveness
Acceptance rates, rework and spend per agent — the evidence for keeping, tuning or retiring one.
Studio Configuration 5
Set the trust level, the models and the isolation boundary.
Capabilities
Turn Studio capabilities on and off to match how far you want agents to go.
Model Config
Connect AI providers. Keys are encrypted with a dedicated AWS KMS key, held server-side and never reach the browser.
Trust Settings
Decide how much an agent may do before a human intervenes — from proposing a plan only, to running verified changes end to end.
Security
Keep Studio data, context and runs inside their own boundary.
Studio Settings
Name, purpose, membership and the Organisation the Studio belongs to.
The other way of working
Productivity Workbench also offers the Product Workspace — team-led agile delivery with optional AI assistance.
Ready to get started?
Create a free account and start with the way of working that fits your team.