You agreed on a project deadline last week. Today, an agent is preparing the next steps. The date is in a conversation, along with two earlier dates you rejected.
It needs to know which decision still applies.
AI agent memory is saved knowledge that an agent can use in later work. In d5s, projects hold memory alongside instructions and files. Your team can keep agreed decisions there, then update them as the work changes.
Here is how to keep rollout decisions useful across tasks.
When you return to a project, you should not have to explain every previous decision again. Glean’s 2026 survey of 6,000 digital workers reports 2.3 hours a week spent supplying AI with context. Dropbox’s smaller October 2026 conference survey found that respondents spent an estimated 36.6% of their AI time finding and bringing information into the task. In d5s, project memory keeps agreed decisions alongside the files and instructions used for later work. You can update what has changed and keep the relevant background available. That can mean fewer repeated searches and explanations each time the project moves forward.
1. Keep the decision, with the context that explains it
Useful memory tells the agent what the team agreed and where that agreement applies.
For a rollout, your team might agree to start with operations and add sales later. Save that agreed scope with its source.
“Start with operations” is worth remembering. Keep the customer name, the agreed scope, and a link to the source with it. Otherwise, a useful decision can become a vague instruction that appears to apply everywhere.
Ask the agent:
Review the scope document and our rollout discussion. List the confirmed decisions worth keeping as project memory, with a source for each. Keep proposals separate. Show me the list before saving it.
Check the list, then ask the agent to save the agreed entries. This review is part of the example workflow. It is not an automatic approval step for every memory change.
2. Give your team shared memory in a project
Project memory lets related work start from the same decisions.
In d5s, create a project when several people or recurring tasks need the same context. Add the source files and instructions, then keep useful decisions in its memory. You can also use Save as project to keep an existing standalone chat and its files in a named project.
The different parts serve different purposes:
| Context | In the rollout example |
|---|---|
| Instructions | Check agreed scope before preparing launch work. |
| Memory | The first phase covers operations, with the project lead as review owner. |
| Files | The scope document and rollout plan. |
| Chat history | The discussion of the options and the work already done. |
| Skills | A reusable procedure for preparing a launch-readiness check. |
Project files and memory are available across work in that project. Conversation history and chat uploads remain with the individual conversation. An agent can retrieve relevant earlier conversations when access permits.
See the d5s project guide for how this context fits together.
3. Update AI agent memory when a decision changes
Keep the current agreement clear without turning a proposal into a fact.
The project lead may hand the rollout review to the delivery lead. Someone also suggests bringing sales into the first phase. Those are different kinds of change: an agreed new owner and a proposed new scope.
Update the owner entry after checking the handover. Keep the scope unchanged until the team agrees otherwise. Add the source and date to the revised note so the next person can check it.
Earlier owner: project lead, kickoff notes, 30 October.
Current scope: operations, agreed scope document.
You can read project memory entries in d5s. People with project edit access can add, edit, or delete them. Check related instructions and files too, so an old owner does not remain in the rollout plan.
Before an important task, review the entries that affect it. Look for an old date, a missing source, or two notes that disagree.
4. Check what the next task carries forward
The test is whether the agent uses the right decision in its next piece of work.
Ask it to prepare the rollout readiness check using the project context and latest progress notes. Ask it to name the decisions it used and flag conflicts.
The result should name the delivery lead as review owner, keep operations as the first phase, and leave the sales proposal open. Compare those details with the source material before acting on them.
In a Project reply, d5s can show searches and reads in the activity. A history lookup can identify the chat and message range read. Use that evidence when checking earlier work, rather than assuming every old conversation was searched.
Our agent testing approach follows the same principle: inspect the result and the evidence behind it.
Questions and answers
Is AI agent memory the same as chat history?
No. Chat history records a conversation. Project memory holds knowledge for later work in that project. Both can provide context, but a saved decision is easier to inspect than a long discussion of possible options.
Can my team share AI agent memory in d5s?
Yes. Project memory supports shared work within the project's access settings. With its files and instructions, it acts as a company brain for that work. Projects inherit workspace access by default and can use Restricted access. Choose an audience appropriate to the information you keep there.
Can I correct or delete a memory?
Yes. People with project edit access can add, edit, and delete project memory entries. Deleting a chat does not delete project memory derived from it, so review that memory separately when removing stored context.
