Discovery takes a Goal or Initiative and asks "How might we…?" — taking an idea from a hunch to reality. You can just ship code to test it, ideate in Figma, whiteboard with your team, or pull the AI into chat to help you think. Whatever the surface, Discovery converges into an aligned Brief that crosses into Delivery when the team's ready to ship.
In Discovery, you turn ambiguity into clarity: a concrete slice of work, tied to a Goal, with acceptance everyone can see. The work lives across conversation, drafts, code spikes, and design exploration — there isn't one dedicated Discovery artefact yet. What follows are the Hamster surfaces that carry it.
Hamster Chat is a persistent canvas where you talk to the AI assistant and your team at the same time, with the Context Graph doing the lookup behind every reply.
Describe a hunch — "We're hearing first-time buyers drop off at the checkout step. What do we know about that?" — and Hamster pulls in the relevant Blueprint, recent Linear tickets, customer-call mentions, prior Briefs, and whatever else is connected, then replies with structured thinking. You ask follow-ups, push back, request a deeper pass, pull in a teammate. The thread is the workspace; the AI is grounded in your team's real systems, not generic.
This is the part most teams underestimate. Chat isn't a wrapper around an LLM. It's the place where ambiguity gets sharpened against everything your team has accumulated, in one room, with the AI as a participant. When the picture is clear, you ask Hamster to turn the conversation into a Brief, and the first draft already references what the chat surfaced.
Research Agents are a deeper, structured Discovery pass for when a quick chat isn't enough. You hand off a question — "Map our last six churn-cohort customer interviews against the activation funnel" — and the agent does the legwork across the Context Graph and any external sources you allow.
Use them when you're scoping an Initiative, validating an assumption, or trying to understand a customer signal across the data your team has already accumulated. The output lands as a structured document you can attach to a Brief as context, so the work the agent did stays connected to the work that ships.
A Brief starts as a hunch in Discovery and stays here while it's being refined. Scope clarifies, the Blueprint comes along as context, alignment votes get collected. The same artefact later crosses into Delivery, which is why every shipped PR can trace back to the conversation that started it.
Refinement is the rest of Discovery once a draft exists: sharpening scope, clarifying acceptance, gathering input from anyone who needs to weigh in. The AI assistant pushes back when scope is vague; you attach Figma frames, customer recordings, related Briefs, and the AI uses them as it tightens the draft. If refining the Brief surfaces something the Blueprint missed, the Blueprint update is proposed automatically — Knowledge stays current as Discovery happens.
When the Brief is good enough that humans and the AI both know what's being shipped, alignment votes go in, and the Brief crosses into Delivery.
You start a chat: "We're hearing first-time buyers drop off at the checkout step. What do we know about that?" Hamster reads what's connected — the relevant Blueprint, recent customer-call transcripts, related Linear tickets — and replies with structured thinking. You ask follow-ups, push back, request a Research Agent pass, pull in a teammate. When the picture is clear, you ask Hamster to turn it into a Brief. The first draft is grounded in everything the conversation surfaced.
From there, refinement is the rest of Discovery — sharpening scope, clarifying acceptance, gathering input from anyone who needs to weigh in. The AI helps tighten scope and success criteria until humans and AI match on what "done" means.
Discovery ends when the Brief is good enough that humans and AI both know what's being shipped. The team votes the Brief ready, a Plan generates, and the work crosses into Delivery.