How Do I Test a Chief of Staff AI Agent With a Made-Up Inbox?
Five fictional emails, five calendar items and four open loops, with a checklist of what the session page says the brief holds.
Paste the sample inbox from the session page into your Chief of Staff AI agent's chat and ask for today's brief. Every name in it is made up, so you can check the result line by line against the session page's list of what the brief holds, with no real inbox connected.
If you aren't ready to connect Gmail or Google Calendar, you can test with a made-up day instead. The sample inbox on the session page has five emails, five calendar items and four open loops, and every name in it is fictional. Paste it in, ask for today's brief, and check the answer against the list below.
From the recording, 46:08. The next thing to do with our agent is run a test. So another three word prompt.
On October 7, 2026 I described the test in three moves: run it, read the first result, then give feedback (46:08 to 46:47). I didn't mention the sample inbox in the session. The file, sample-inbox.md, gives you something to run the test on.
What is inside the sample inbox?
The file opens by describing itself:
This is a made-up inbox and calendar for anyone who isn't connecting Gmail or Google Calendar today. Every name, company, and email address below is fictional. No real person or organization is represented. Paste the whole thing into your Chief of Staff AI agent's chat and ask for today's brief.
It holds three things: a calendar, five emails and a list of open loops.
The calendar, in time order
| Time | Item | | --- | --- | | 9:30 AM | Team check-in, weekly, 30 minutes | | 11:00 AM | Call with Priya Nandakumar, Halverson Group, proposal review | | 1:00 PM | Lunch hold, no meeting | | 3:00 PM | Intro call with Tomás Reyes-Aguilar, Brightline Partners, referred by Dana | | 5:30 PM | Dentist |
The five emails, and what the session page lists for each
| From | Subject | What it asks | In the brief | | --- | --- | --- | --- | | Priya Nandakumar | Re: Proposal, two questions before our 11 | Whether the second phase is in the number on page 3, and the earliest start date | A drafted reply | | Dana Whitfield | Intro: Tomás at Brightline | Introduces Tomás and leaves the next step to you | A drafted reply | | Lumen Workspace | Your invoice for October is ready | Nothing; it is an automated message | No draft | | Marcus Oyelaran | Following up on the panel | A 20-minute call next week about a fall panel | A drafted reply | | Lena Varga | Quick one | Whether the reading list was ever sent | A drafted reply |
The four open loops, with the ages the file states
| Open loop | Age in the file | | --- | --- | | Reading list owed to Lena | Promised two weeks ago | | Reply owed to Marcus about the panel | Eight days ago | | Placeholder start date in the Halverson proposal | To be fixed before today's call | | Thank-you note owed to Dana for the intro | None stated |
How do you run the test?
The steps come from the session page, not from a screen I opened on October 7, 2026.
- 01
Open sample-inbox.md from the session page and copy the whole file.
- 02
Paste it into the chat of your Chief of Staff AI agent.
- 03
Ask: "Give me today's brief."
- 04
Read the answer against the checklist below.
The instructions file on the session page asks for three parts in a fixed order: today's calendar, emails that need a reply, and open loops. For the last part it says:
Name each one plainly. Don't soften it into a suggestion if it's actually overdue.
What does the session page say the brief holds?
The session page lists what to look for. Each row below restates one item from it.
| Look for | Where it comes from | Present? | | --- | --- | --- | | The calendar in time order, five items | The calendar table above | | | A drafted reply for Priya | Email 1 | | | A drafted reply for Dana | Email 2 | | | No draft for the Lumen invoice | Email 3, automated | | | A drafted reply for Marcus | Email 4 | | | A drafted reply for Lena | Email 5 | | | All four open loops named plainly | The open loops table | | | The overdue loops called overdue | The ages in the open loops table | |
What does a missing line mean?
The session page says that if any of those items is missing, that is your correction for Step 6. The correction is one sentence. The instructions file describes how the AI agent takes it:
If I tell you something is off, in one sentence, update the next answer to reflect it. Don't ask me to explain further unless my correction doesn't make sense. Keep the fix; don't quietly revert to the first version next time.
On October 7, 2026 I also described how feedback sounds when a result is only partly right: you tell the AI agent what you like and what you don't like, and you tell it why (46:32).
If you leave something unsaid, the AI agent may ask about it. During the build I told the group I hadn't said how I wanted the information outside a dashboard, so the AI agent would ask me clarifying questions (13:05).
What goes wrong
I named one of these in the session, and the file names the other:
- A result you do not like. I said to tell the AI agent why in each case: when you love the result, when you have mixed feelings and when you do not like it (46:26 to 46:42). I said not to say "try again."
From the recording, 46:47. So don't say try again, actually say, "Hey, you surfaced this and this isn't beneficial and this is why."
- A fix that does not carry over. The file asks the AI agent to keep a correction and not revert to its first version next time.
What the recording does not show: the sample inbox pasted into a chat, or the brief it returns. The lines to look for come from the session page and the instructions file, not from a screen I opened.
Where this comes from
This article grows from the session I taught live with Women Applying AI on October 7, 2026. The quoted passages are my own words, taken verbatim from the recording or from the two files on the session page. The tables, steps and checklists around them were drafted by my team's AI agents from those words and my published pages, and checked against the recording before publishing.
Build Your First AI Agent in One Hour: the full session with the clips, the prompts and the transcript. The files are on the session page.
Related reading: How Do You Test an AI Agent and Give It Feedback?, How Do You Practice With an AI Agent Without Risking Real Data?, What Does Run a Test Mean for an AI Agent?, How Do I Show My AI Agent to an Audience Without Exposing Real People's Data?.
Frequently asked questions
Is anyone in the sample inbox real?
No. The file says every name, company and email address in it is fictional and that no real person or organization is represented.
Which emails get a drafted reply in the brief?
The session page lists drafted replies for Priya, Dana, Marcus and Lena, and no draft for the automated Lumen invoice.
What counts as a correction if the brief leaves something out?
The session page treats a missing item as your correction for Step 6. Say what is off in one sentence, and the instructions file asks the AI agent to update the next answer and keep the fix.
Build something that actually runs your workflow.
A focused, free 60-minute live session with Jacqueline. You build alongside her, on your own real task, and leave with an AI agent that is already running.