The product · beta # Orient before work. Capture as you go. Let your agent check the context, capture the reason and reuse what your team knows—from the tools you already use. [Start your workspace ↗](https://work.productbrain.io/sign-up)[Try it with a real task →](/get-started/#setup) Where it happens Your usual**agent** **Product Brain** Shared decisions · reasons · roles Read and write through CLI or MCP. Keep working in your connected agent. [01 Orient](#orient)[02 Work & capture](#capture)[03 Reuse together](#reuse)[04 Learn & refine](#refine)[05 Connections](#practice) Orient before work ## Your agent finds what applies. What exists, what’s decided, what came before—and who can change it. Your usual work request “Can we include this extra workflow in the proposal?” With the practice configured, your agent reads its role and relevant records. Partial orientation needs further lookups before acting. Walkthrough · a customer request ### An extra request. An existing agreement. Current · accepted decision #### Keep the pilot within the agreed scope. **Why** · Deliver the first result before adding custom work. **Who can change it** · Product Lead Direction Prove value before adding complexity. Work Prepare a customer proposal. What exists? What came before? **Already there** · The current offer and pilot plan. **Earlier choice** · A broader pilot. Its reason—exploring more possibilities—stays in history after the team narrows the scope. Your agent’s next step Check with Product Lead before promising the extra workflow. An existing agreement informs the next action. Work & capture ## Your agent keeps the reason. A new choice emerges during work. Your agent records its reason, scope and owning role. As you work together “Let’s test the extra workflow with one customer before offering it more widely.” Your agent can capture and accept in its acting role. You remain accountable. Accepted isn’t delivered. Walkthrough · a new choice during work ### One customer. One scoped experiment. 01 · Draft **Record the choice.** Why: learn what it takes before adding it to the offer. 02 · Accepted **Establish the agreement.** Product Lead—or their agent in that role—accepts. 03 · Work done **Record delivery.** The experiment runs. Record what happened. What must be in place to accept? The agent’s session must act in the role that can decide. Holding another role does not add its authority to this session. Required fields and applicable direction links, limits and verification requirements must also be met. Agents can accept any entry kind within those requirements. Acceptance is recorded for the person they act for. The person remains accountable and can confirm it themselves. Saving a draft, accepting knowledge and completing work are different steps. Your working preference ### Choose how your agent captures. Set this in its working instructions. #### Show me before saving. Think it through together, then record. #### Capture as we go. Delegate the record within your role. Both follow your team’s rules and role authority. You remain accountable. [Give your agent the working instructions →](/get-started/#agent-instructions) Reuse together ## The next person starts with the why. Sales asks their agent to prepare the next proposal. It can consult the same decision and reason—even while the author is away. Original context holder ### Product Lead Away from the work Accepted decision ### One customer. One scoped experiment. Reason · Learn what it takes before adding it to the offer. Owning role + reasons + history Sales + agent Set expectations from the agreed offer Delivery + agent Plan the agreed scope Marketing + agent Describe the actual offer Shared across roles, even while the author is away. Capture Connect Review Reuse Refine ↻ Knowledge compounds when your team captures, maintains and reuses it. That’s our thesis. Learn & refine ## New learning. Your agent brings it back. New learning prompts review. The agreement stands until its owning role changes it. Through your agent “Customers want this workflow. Delivery needs more time. Let’s revisit the offer.” Possible contradiction flags prompt review. The owning role—or its agent in that role—decides what changes. 01 · Current agreement #### One customer. One scoped experiment. Why · Learn before adding it to the offer. 02 · New learning #### Useful for customers. More work to deliver. Candidate: an optional add-on. Not yet accepted. 03 · Review #### Does the add-on fit our direction and capacity? Check the learning against the current agreement. 04 · If the owner agrees #### Optional add-on. Separately scoped. The owner records why. Earlier versions stay in history. Walkthrough · New learning → review → authorised change. ### Decisions stay close to the work. Your team defines the scope. The role holder—or their agent in that role—accepts and changes the agreement. What if it isn’t your role? Attempt a change **Change refused** The response names who decides. Make the issue visible **Flag the issue** Record it as a tension—an issue to resolve—or ask the owner. Owning role **Decide the change** The authorised role changes the accepted record, with history. There isn’t an amendment-proposal workflow for an already accepted entry. The candidate above is an issue for review, not a queued edit awaiting approval. No owner or an unfilled role? Team Lead fallback applies. [See the practical answer →](/get-started/#who-decides) Across the team ## Keep using the agent you know. Connect a supported CLI/MCP agent and teach it your practice. Day-to-day lookups, capture and authorised acceptance can stay there. Across functions ### Claude web + desktop Hosted MCP connector. Look up your workspace’s decisions, reasons and related work. [Claude connection guide ↗](https://productbrain.io/docs/integrations/claude) Connector · beta MCP-enabled plans ### ChatGPT Custom MCP connectors on supported plans. Check your plan and rollout. [Check the ChatGPT guide ↗](https://productbrain.io/docs/integrations/chatgpt) Conditional connection · beta Agent-led development ### Coding agents Claude Code: opening briefing. Cursor and Codex: CLI or MCP lookup. Copilot: instruction files. [Choose your agent guide ↗](https://productbrain.io/docs) Claude Code · Cursor · Codex · Copilot · beta Teach your agent to consult the record as it works. Changes aren’t pushed to every active session. [Start your first lookup →](/get-started/#setup)[Share your practice in the community ↗](https://community.productbrain.io/) ## Build on what your team already knows. Give your connected agent a real task. Let the next person build on the recorded reason. [Start your workspace ↗](https://work.productbrain.io/sign-up)[See the setup steps →](/get-started/#setup)