Create your first read-only Agent
Start with an Agent that reads information and returns a proposal without changing a Room or sending anything. This guide is for designing, evaluating, and deploying a reusable Agent; it is different from asking the Room AI Agent a one-off question.
Example job
Create an Agent that reads meeting notes, lists decisions, follow-ups, and open items, then proposes agenda items for the next meeting.
- Select 新しいAgent (New Agent) in Agent Studio.
- Describe the goal in the ジャーニー (Journey) input.
- If needed, add the meeting notes from the target Room as material.
- Select 設計案を作る (Create design proposal) and check its goal, sources, output, and boundaries.
- Select この設計で下書きを作成 (Create draft with this design), then continue to シミュレーション (Simulation).
Example request:
From these meeting notes, organize decisions, follow-ups, and unresolved items, then propose next-meeting topics in priority order. Do not update the Room or contact anyone externally.
Make a request easier to check
| Include | Example |
|---|---|
| When it runs | After a meeting, every morning, when a file is added |
| What it reads | Meeting notes, Room tasks, deal summary |
| What it returns | Decisions, next actions, an executive summary |
| What it must not do | Do not update, send, or run without review |
Safe first configuration
For the first version, confirm all of the following:
- Scope: one selected Room only
- Run: manual runs only; no schedule or webhook yet
- Connections: no MCP action beyond reading
- Output: proposals, summaries, and drafts only; no Room update or external delivery
Only add a trigger or permission after the manual result is consistently useful.
Words such as “send” or “update” can add a write or external action to the design. Start with “propose” or “prepare a draft” and evaluate the result first.
Related articles
- What is Agent Studio?Understand how to describe a job, then create, evaluate, deploy, and improve an Agent
- Evaluate, approve, and deployCheck expected results in simulations before deploying an Agent to a Room
- Rooms, sources, connections, and permissionsUnderstand what an Agent can reference and what to check before enabling MCP or write actions