AI Agent Daily Brief: Set One Up in 20 Minutes
Your daily brief is the highest-value thing a personal agent does. What to put in it, the exact prompt and schedule, and what to leave out of it.
A daily brief is one scheduled message from your agent, waiting before you start your day: what is on your calendar, what needs a decision, what changed overnight, and what it already handled. It takes about twenty minutes to set up, and it is the single highest-value thing most people ever get out of a personal AI agent — because unlike a chat you have to remember to open, a brief arrives on its own.
This is the complete version: what belongs in it, the prompt to use, where to schedule it, and the failure modes that make people turn briefs off.
What belongs in a brief
Five blocks, in this order. If a block has nothing, it says "nothing" and moves on — no filler.
- Today's shape. Calendar for the day, with travel time between back-to-back items and any conflict flagged. This is the one block worth getting exactly right.
- Needs you. Messages where someone is waiting on a reply or a decision — a school email, a client question, an invoice with a due date. Not "you have 34 unread emails"; the three that matter.
- Due or drifting. Tasks and renewals with a date: bills to pay, forms to sign, subscriptions renewing, anything with a deadline in the next 14 days.
- Watch list. The handful of numbers or pages you care about — a price you are waiting to drop, a package, a flight fare, a metric. Only what changed, with the change.
- Done. What the agent completed since yesterday. Short, past tense, with the artifact where you can check it.
Notice what every block has in common: it is read-only or already-finished. Nothing in a brief spends money, sends anything, or makes a commitment. That is what makes it safe to run at 06:30 while you sleep.
What to leave out
- The raw inbox. A dump of subject lines is noise with a timestamp. Triage, don't forward.
- Anything the agent cannot verify. If a calendar feed fails to load, the correct brief says "calendar unavailable", not a confident empty day.
- Anything outward-facing without review. Drafts belong in the brief; sends do not.
- Motivation and commentary. You do not need a paragraph about how your week looks. You need the three things.
- More than one screen. If it does not fit on a phone screen, it will not be read by Thursday.
Set it up: five steps
- Pick three sources. Calendar plus one inbox plus one more (tasks, notes, a specific site). Adding sources is easy; trusting a brief that quietly skipped one is not.
- Write the prompt with explicit gaps. Tell it what to say when data is missing — this is the difference between a brief you trust and one you stop reading. Template below.
- Schedule it 30 minutes before you wake. Early enough to beat you, late enough that overnight emails are in. A weekday schedule plus a lighter weekend version works better than one job doing both.
- Deliver it where you look. A push notification to your phone and the message sitting in a conversation beats an email you will find at lunch.
- Review it daily for one week, then cut. Every block you skipped reading twice gets deleted. Most briefs fail at nine blocks, not four.
The prompt
Every weekday at 06:30, build my daily brief from my calendar,
my inbox and my task list. One screen. Five blocks, in this order:
1. TODAY — calendar items with start times and locations. If two
events are back to back, note travel time. Flag conflicts.
2. NEEDS ME — max 3 messages that need a reply or a decision
today, with who and what they want in one line each.
3. DUE — anything with a deadline in the next 14 days, plus
subscriptions renewing this month.
4. WATCH — only items whose value changed since yesterday.
5. DONE — what you completed since the last brief, with the file
or link where I can check it.
Rules: if a source fails to load, write "unavailable" for that
section and never guess. No advice, no commentary, no summary
of the summary. Ship nothing, send nothing, buy nothing.
If your agent supports persistent facts about you, add a line telling it who the important people are — "anything from school or my accountant goes in NEEDS ME". A brief improves fast once it knows your priorities rather than guessing them.
The three failure modes
| Failure | What it looks like | Fix |
|---|---|---|
| Confident emptiness | "Nothing on today 🎉" when the calendar feed was actually down | Require an explicit "unavailable" state per source |
| Volume inflation | Twelve bullets, all equally urgent | Hard cap every block (3 items max), let the agent rank |
| Alert fatigue | A ping for something that could have waited | Reserve high priority for money, travel and hard deadlines |
The third one is why briefs die. If the notification goes off on your phone for something that mattered on Tuesday, you will mute it by Friday — and it will not come back.
Make it a system, not a one-off
A brief is the natural first automation because it is read-only, bounded, and immediately useful. From there it becomes the base layer other jobs report into: the subscription check writes to DUE, the price watch writes to WATCH, the inbox triage writes to NEEDS ME. One message a day, several agents behind it.
Two things to get right before you add more:
- Schedule what is safe to leave alone. Read-and-report first; anything that acts gets a review step. What to schedule unattended has the checklist we use.
- Set it up in your own stack. Scheduling a personal agent's automations covers the mechanics, and what to automate first helps you pick the next job after the brief. If you want the brief itself ready-made, Wolffish ships scheduled briefings with review gates as a first-class flow.
CC on families shows where the pattern is heading — Google's "Your Day Ahead" is the same idea, sent to six people instead of one. Whether it is you or your household, the design rule holds: one screen, ranked, honest about what it could not read.
The takeaway
Twenty minutes, three sources, five blocks, one schedule. A daily brief is the cheapest way to find out whether an agent fits your life — because it asks nothing of you but a glance, and it fails loudly rather than silently. Build it this week, read it for seven days, and delete whatever you skipped. What survives is the automation you actually wanted.
