How an agent can use a decision from an earlier session
Artur Sabirov2 min readRevised September 7, 2026. Product details reflect this revision.
Suppose your team decides to keep the old billing system running until a migration is finished. The agent helping with the migration knows why: some customers still need invoices from the old system.
Next week, another agent reviews the code and suggests removing the old billing integration. If the earlier decision is missing from its context, it has no reason to keep that integration.
You can paste the decision into each new session. We spent months doing that with company context. Brain gives agents a place to store those decisions and search for them later.
Save the decision and its reason
In this example, an agent could store:
Keep the old billing integration until the migration is complete. Customers who have not migrated still need it for invoices.
The reason matters. “Keep the old integration” records an instruction. Explaining who still needs it also helps the next agent understand when that instruction might stop applying.
Brain keeps the submitted text and extracts individual statements as searchable memories. Each product has its own memory pool.
Find it when the work needs it
When another session starts work on billing, the agent can search Brain for relevant decisions. The stored explanation gives it context for evaluating whether the old integration can be removed.
Agents read and write Brain through MCP. For a decision other agents need, they can use shared memory. Agents with Brain access in that product can read it. An agent can also keep private working notes that only it can read.
Storing a decision makes it available. The agent still needs to retrieve it and use it in the task; memory does not mean every saved fact appears in every answer.
Update it when the decision changes
Now suppose the migration finishes. The old instruction should no longer guide new work.
A new statement can replace an older one when it has the same or higher source authority. If it has lower authority, Brain records the conflict for clarification instead. Lists and searches show the current version.
That makes the reason in the original memory useful again: the condition for keeping the integration has changed. The team can record the new decision and why it changed.
Make time to check what is stored
A stored mistake can mislead later work just as an accurate decision can help it. You can list memories, inspect their sources, correct their content, and delete entries that should not be there.
Start with a decision your agents repeatedly need explained. Store the decision with its reason, then check what an agent finds when it searches for that subject. That gives you something concrete to review before relying on memory for more of your work.
Related posts
Building Lessly in the open
Follow along, and get early access to the private beta.