Core Concepts

Understanding these concepts will help you use the AI Product Studio confidently and communicate clearly with your team about AI-assisted delivery.

Intent

A unit of work captured in natural language — a problem, idea, bug, spike, or outcome statement. Unlike traditional tickets, intents are deliberately vague at capture; shaping adds structure.

Kinds: opportunity, bug, spike, tech-debt, idea, outcome

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

Agent

A configured AI worker specialised for a specific type of task. Nine types: Composer, Coder, Reviewer, Security, Tester, Researcher, Compliance, Accessibility, and Performance. Each has configurable trust level, cost cap, and capability set.

Plan Gate

Before an agent executes, it produces a step-by-step plan. A named human must read and acknowledge this plan. Plans can be rejected — the agent will re-plan. The Plan Gate is the most cost-effective place to catch misunderstandings.

Context Library

A curated set of knowledge artefacts (ADRs, coding standards, personas, compliance controls, API contracts) that agents are grounded in. Context quality is the biggest driver of agent output quality.

Review feed

Everything the Definition of Done’s agents want from each pull request, across every package in delivery: rework, questions to answer, stale evidence and merge holds. Every assessment, answer and decision is kept on the work package as append-only evidence.

DPIA

Data Protection Impact Assessment. Documents what data is processed by AI models, for what purpose, under what controls. A release containing AI-generated changes can be approved only when it names an approved DPIA.

Verify

What the Definition of Done found across every package in delivery — by concern, by pull request next to your own CI, and finding by finding.

Outcome Telemetry

Records the metrics and signals each release is judged by. With the Feedback Loop, it shows whether a release moved the key results it named — not just that it shipped.