Intent that survives a handoff
Switch AI. Keep the work.
Save the purpose and constraints. The next person or AI continues from there, tomorrow, after a review, or in a different tool. What worked can become a Playbook.
Case · Acme renewal
Work that can be resumed
Researcher
Claude CodeFinished the research and left the evidence, assumptions, and open questions in the Case.
Account exec
ChatGPTNowPicks up the thread on a different AI and keeps going, without redoing the research.
Enterprise renewal review
After the Case succeeds, this way of working can be reused: which sources count as evidence, and what a person has to check.
01 · Continue the work
Carry the work across people, AI, and tools.
Epismo does not copy every tool call or thought. It keeps the state collaborators actually need: the matter at hand, explicit responsibility, and durable outcomes. That is enough for the next collaborator to understand what happened and continue with confidence.
Case
One real piece of work, pinned to a Playbook Version or started ad hoc. It can be lightweight and finish without Tasks.
Task
Materialize responsibility only when someone needs to own a piece of work, review an outcome, or accept a handoff.
Record
Append outcomes, decisions, notes, and references so another collaborator can resume without replaying the entire session.
Then reuse
When it works, turn it into a Playbook.
The first value is continuing the work in front of you. A Playbook is how a successful Case becomes a method the next Case can follow.
Enterprise renewal review
How this team assesses a renewal: which sources count as evidence, and what has to be checked by a person.
- 1Scope the renewal risk this research has to answerSkill·Renewal risk rubric
- 2Pull filings, support history, and usage into one evidence setMCP·Filings & earnings callsMCP·CRM + ticket history
- 3Separate what is verified from what is still assumedPlugin·Claim-to-source auditCLI·Exec brief builder
- 4Hand the recommendation to the account ownerTask · Review
Every Step names the Skills, MCP servers, CLIs, and plugins it expects, so a teammate's agent reproduces your run instead of improvising its own.
02 · Playbooks
Make the way of working reusable, without freezing the work.
A Playbook captures durable guidance: what to consider, which resources may help, and what a good outcome looks like. The agent runtime still decides how to execute.
Guidance, not a rigid graph
Agents can skip, combine, reorder, or extend Steps as the situation requires.
Runtime-agnostic
Use the same Playbook from Codex, Claude Code, an MCP client, or your own agent.
Auditable after the fact
What you published never changes, so you can trace which steps a past piece of work actually followed.
A learning loop
Turn lessons from real work into Suggestions, and improve the next run without rewriting history.
The loop
Save once. Continue. Reuse what worked.
Keep intent in the Case, continue from another person or AI, and turn a successful run into a Playbook when it is worth repeating.
Save
Keep purpose, constraints, and current decisions.
Improve
Lessons from real work feed the next run.
Continue
The next person or AI picks up without a restart.
Reuse
A successful Case can become a Playbook.
Start with one continuation
Stop making every AI start from zero.
Save the work once, continue it with the next person or AI, and let a successful Case become a Playbook when you want to repeat it.