Quick answer
To brief a chief of staff AI agent on when to interrupt you, give it a short written brief with three parts: your top priorities for the next few weeks, a narrow list of things that may interrupt you straight away (missed commitments, new evidence that changes a decision, choices only you can make), and a rule that everything else waits for one daily digest. Ask it to rank honestly, say when something is low value, and explain what each decision would cost you in time.
Key takeaways
- A chief of staff agent should protect your attention, not fill it
- Keep the interrupt list short and specific
- Everything else goes into one daily digest at a set time
- Ask it to say plainly when something does not matter
- Review and tune the brief weekly for the first month
In this guide
- What is a chief of staff AI agent?
- Why does it need a brief on when to interrupt you?
- Step 1: What are your current priorities?
- Step 2: What is allowed to interrupt you?
- Step 3: What can wait for a daily digest?
- Step 4: How do you stop it calling everything important?
- Step 5: What does a complete brief look like?
- Step 6: How do you tune it after the first week?
- Frequently asked questions
- The bottom line
- Related posts
This guide supports our main article, AI Agents and Productivity Theater: 4 Principles to Make Them Do Real Work. In it, Nate B Jones argues that the agent managing your work needs a “budget for your attention”. Here is how to write that budget down as a practical brief.
📘 Main article: Read AI Agents and Productivity Theater: 4 Principles to Make Them Do Real Work for the four principles behind this guide, including why more agent output often means more work for you.
What is a chief of staff AI agent?
A chief of staff AI agent is an assistant that coordinates your other AI tasks and decides what to bring to you. It might summarise what your agents produced overnight, track follow-ups, or flag decisions you need to make.
You can set one up in most capable AI assistants that support custom instructions, projects or scheduled tasks. The tool matters less than the brief you give it.
Why does it need a brief on when to interrupt you?
Because agents have almost unlimited attention and you do not. Without a brief, a manager agent tends to pass on everything that looks important, and your morning becomes a long, neatly organised list of demands.
A good brief turns it into a filter. It knows what you care about, what you are already committed to, and how much a new idea would really cost you to pursue.
⚠️ Warning: Do not give a chief of staff agent permission to send messages, spend money or accept commitments on your behalf without approval. Its job is to prepare decisions, not make them for you.
Step 1: What are your current priorities?
Write your top two or three priorities for the next four to six weeks. Be specific:
- “Launch the new mini-course by 30 November”
- “Grow the email list with one lead magnet”
- “Keep existing clients happy; no missed deadlines”
Add your existing commitments, such as client work, hiring or learning a new skill, so the agent can judge what a new task would add to your load.
Step 2: What is allowed to interrupt you?
Keep this list short. Three to five items is plenty. Based on the principles in Nate’s video, good candidates are:
- A commitment at risk: a deadline, promise or follow-up about to be missed
- Evidence that changes a decision: new information that contradicts something you have already decided
- A decision only you can make, with a real deadline
- Something tied to a stated priority that is unusual or new, such as an unexpected customer comment
💡 Tip: For each interrupt, ask the agent to explain what changed, give you the options and say what each option would require from you. That makes the interruption quick to act on.
Step 3: What can wait for a daily digest?
Everything that is not on the interrupt list goes into one daily digest, at a time you choose, such as 9am. Ask for a fixed format:
- Decisions needed (most important first, with a recommended option)
- Progress on your priorities, in one or two lines each
- Experiments still running and whether they are teaching you anything
- Suggested stops: tasks that could be ended
- Parked ideas: interesting but not for now
The “suggested stops” section pairs well with our companion guide, How to Set a Stop Condition for a Recurring AI Agent Task.
Step 4: How do you stop it calling everything important?
Tell it directly that honesty about low value is part of the job. Nate suggests a “truth-teller” style instruction, so agents say “this does not matter much” or “this is only an experiment” instead of labelling everything urgent. Here is our own simple version you can adapt:
“Be honest about importance. If something is low value, say so in one line. If an item is only an experiment, label it as an experiment. Do not call anything urgent unless it matches my interrupt list. When unsure, put it in the digest, not an interruption.”
Step 5: What does a complete brief look like?
This is an example brief to adapt, not a setting from a specific tool.
| Section | Example wording |
|---|---|
| Role | You are my chief of staff. Your job is to protect my attention and prepare decisions, not to make them. |
| Priorities | 1) Launch the mini-course by 30 November. 2) Grow the email list. 3) No missed client deadlines. |
| Commitments | Two client projects; recruiting a part-time assistant this month. |
| Interrupt me only for | Deadlines at risk; new evidence that changes a decision; decisions only I can make with a deadline under 48 hours; unusual customer feedback. |
| Everything else | One digest at 9am in this order: decisions, progress, experiments, suggested stops, parked ideas. |
| Honesty | Say plainly when something is low value or only an experiment. |
| Limits | Never send, spend or commit on my behalf without my approval. |
💡 Tip: For a fuller look at where to draw approval lines for AI helpers, see How to Set Human Approval Boundaries for AI Workers.
Step 6: How do you tune it after the first week?
Check how many interruptions you had and whether each one was worth it. At the end of the first week:
- If more than one or two interruptions a day were not worth it, narrow the interrupt list
- If something important was buried in the digest, add it to the interrupt list
- If the digest is too long to read in five minutes, ask for a shorter format or fewer sections
- Update your priorities when they change, so the agent is not working to an old plan
📘 Main article: Want the bigger picture? AI Agents and Productivity Theater: 4 Principles to Make Them Do Real Work explains all four principles, and our stop-condition guide helps you clear out agents you no longer need.
Your next step: write your three priorities and a five-item interrupt list today, paste them into your AI assistant’s custom instructions, and judge the results after one week. Then read our main article on AI productivity theater.
Frequently asked questions
Which AI tool should I use as a chief of staff agent?
Any capable assistant that supports custom instructions or projects can work. Start with the tool you already use and focus on the brief rather than the software.
How often should a chief of staff agent contact me?
Aim for one daily digest plus rare interruptions. If you are being interrupted several times a day for things that could wait, the interrupt list is too broad.
Can a chief of staff agent manage my email?
It can summarise and flag emails if you connect it, but keep sending and replying behind your approval, especially for customers and suppliers.
What if the agent misses something important?
Add that type of item to the interrupt list and give an example. Briefs improve quickly when you correct them with real cases.
The bottom line
A chief of staff agent is only as good as its brief. Tell it your priorities, keep interruptions rare and specific, push everything else into one digest, and ask for honesty about low-value work. Your next step: write your interrupt list today and test it for a week.
Related posts
- AI Agents and Productivity Theater: 4 Principles to Make Them Do Real Work
- How to Set Human Approval Boundaries for AI Workers
- Claude Duet: Anthropic’s Proactive AI Agent, Plus Gemini 4 Carbon’s New Checkpoint
