Build a Personal AI Chief of Staff (7am Briefing)

My chief of staff is an AI named Audrey Two. (Yes, like Little Shop of Horrors.) Every morning at 7am my briefing lands in Telegram, and by the time I'm out of bed I already know what I'm walking into.

This is the assistant I tell everyone to build first, and this is the full setup.

(The one-sentence version, for the skimmers: an AI daily briefing is a standing instruction that makes an assistant assemble your day, calendar, email, tasks, weather, into one short scan that's ready before you've opened a single app.)

What it's connected to

Three connections, and deliberately not more:

My calendar
My email, both my work inbox and my personal one
ClickUp, where all the project work lives

Notice what's missing: the team accounts. My team manages the shared inboxes, so my assistant stays out of them. The rule is that it reads what's yours to read, and only that. Whatever your version of a shared inbox is, leave it out.

What the briefing covers

Mine is a morning scan, a quick pre-triage of the day:

What's on my calendar, with a flag on anything complex or back-to-back (and reminders to do things like, say, eat lunch)
Where my deep work block should go, and whether I have external meetings
Anything urgent sitting in either inbox that needs a reply
The weather, a daily news and content briefing, and any reminders. What's left, what's blocked.

None of that's information I couldn't find myself. The point is that I don't have to go find it across five apps before coffee.

The instruction that makes it work

Here's a briefing instruction you can adapt. It goes in your assistant's standing instructions, not typed fresh each day:

Every morning at [7am], send me my briefing. Include: what's on my calendar with a flag on anything complex or back-to-back, where my deep work block should go, whether I have external meetings, anything urgent in my inbox that needs a reply, the weather, and any reminders. What's left, what's blocked. Remind me to eat lunch on packed days. This is my pre-triage: I want to know what I'm walking into before I get out of bed.

Adjust the list to your real morning questions. The test for whether something belongs in the briefing is simple: is it a thing you'd otherwise check on your phone before getting up? Then it goes in. Everything else is noise at 7am.

What a good briefing reads like

A template of the output you're aiming for, so you can correct yours toward it:

Good morning. Here's your day.

CALENDAR: 4 meetings. 11 to 1 is back-to-back with no break, and your
2pm is external (prep notes are in the project). Nothing after 4.

DEEP WORK: Your best block is 8:30 to 10:30. Protect it; tomorrow is
worse.

INBOX: Two things need replies: [sender] on the contract question, and
[sender] asking to move Thursday. Everything else can wait.

WEATHER: 84 and humid. The 6pm walk is the good one.

REMINDERS: Invoice goes out today. You have no lunch gap; eat at 1.

STILL OPEN: The proposal draft is blocked on you. Nothing is blocked
on anyone else.

Short enough to read in bed, specific enough to act on, and it leads with the thing that will actually shape the day. When yours comes back as a wall of pleasantries and bullet soup, correct it against this shape: lead with the calendar risk, keep every line actionable, cut anything you already knew.

How to build it

Two paths, and the difference is whether the briefing arrives on its own.

The evening version lives in one project with one instructions file. No connections. You paste in your calendar and your inbox highlights each morning, and it does the triage. Twenty seconds of pasting, same brain. The one-file assistant guide covers the whole setup, including the interview prompts that write the instructions for you.

The built version connects to your actual calendar, email, and project tool, and runs on a schedule so the briefing is waiting when you wake up. That's a weekend project in Claude Code, covered start to finish in the full build guide, including the part almost everyone skips: writing down what it may do alone and what it must never touch, before you connect anything.

Start with the evening version. If you find yourself pasting the same three things every morning for two weeks, that's your sign to build the connected one.

Tonight's version, in 20 minutes

If you want this working tomorrow morning:

Create a project (Claude or Cowork) called something like "Chief of Staff."
Run the interview prompt from the one-file guide and let it write its own instructions. Add the briefing instruction above to them.
Drop in one context file: how your week normally runs, your standing commitments, what counts as urgent for you, and what never does.
Tomorrow morning, paste in your calendar and anything on your plate, and ask for the briefing.
Correct whatever's wrong, out loud, and move each correction into the instructions.

That's the whole first version. The scheduled one can come later, or never.

When the briefing comes back wrong

The four standard failures, and the fix for each:

It's too long. Add "the whole briefing fits on one phone screen" to the instructions. Length is a setting, not a personality.
It buries the urgent thing. Add an ordering rule: calendar risk first, then what needs a reply, then everything else.
It tells you things you already know. Add "skip anything I'd already know from a glance at my calendar; your job is the flags, not the inventory."
It invents calendar items or emails. That's a context problem, not a tone problem. In the paste version, it only knows what you gave it; in the built version, check the connection before assuming anything else.

What changes

The briefing doesn't do my work. It sequences it. I stop reconstructing my day from five apps, my deep work block gets placed instead of squeezed, and the urgent email gets answered before it becomes an urgent phone call.

And on the days that are wall-to-wall, it's the thing that reminds me to eat lunch.

Common mistakes

Connecting everything on day one. Calendar, email, and one work tool covers almost all of it. Every extra connection adds noise to the scan.
Letting it into shared or team accounts. Your briefing is about your day. Team surfaces belong to the team.
Making the briefing exhaustive. A pre-triage you stop reading is worse than none. If you skim past a section three days running, cut it from the instructions.
Treating the first week's output as final. Correct it in plain language and move each correction into the instructions, where it becomes permanent.

FAQ

Does this need to be its own assistant, or can my general chat do it? Its own. A briefing depends on standing instructions and consistent context, and a general-purpose chat holding your whole life gets worse at all of it. One assistant, one job.

What tool does this run in? The evening version works in a Claude Project or Cowork. The scheduled version is a Claude Code build. ChatGPT has its own equivalents of both.

What about my to-do list app? Connect the one where work actually lives and skip the rest. An assistant reading an abandoned task manager reports fiction.

Can it send the briefing somewhere instead of living in a chat? That's the built version. Mine arrives in Telegram on a set schedule. Wherever you already read messages first thing is the right delivery.