In the Workshop: Part 2

In Part 1, we picked the three things you need: a Mac, an AI model, and an agent system. Now we make the machine useful.

The daily brief is the hello world of agent automation. It is small enough to set up in an afternoon, but useful enough that you'll miss it when it breaks.

This is the smallest workflow that proves the always-on Mac thesis. A Mac sitting on a shelf is not interesting by itself. A Mac that quietly checks your inbox, calendar, tasks, and team messages before you wake up is the point.

A quick setup note: my headless machine is an old MacBook sitting on a shelf with the lid closed. When I need to configure it, I use Astropad Workbench to remote into it from my main computer. It feels like sitting in front of the machine, without dragging it out of its little cave.

Install Claude on the Mac that will do the work, not your main machine.

Cowork lives inside Anthropic's Claude desktop app. Grab the macOS installer, drag it into Applications, sign in, and make sure you're on Claude Pro or Max. The free tier does not include Cowork.

Once you're signed in, you'll see the sidebar: Chat, Code, Cowork. Click Cowork. That's where we'll spend the rest of this issue.

If you're running this on a headless Mac, do this setup through remote desktop. In full disclosure, we make Astropad Workbench, which is what I use for this. Any reliable remote desktop setup works, but you want something you trust because you'll be configuring accounts and connectors.

The agent can't help until it can see the things you care about.

Without connectors, Cowork is just a very polite empty room. It can't see your inbox, your calendar, your Slack, or your task list. So before touching the scheduler, wire up the data sources.

In Cowork, click Customize, then Connectors. You'll see the usual suspects: Gmail, Google Calendar, Google Drive, Slack, Todoist, Microsoft 365, Linear, Notion, Zoom, Figma. Pick the ones that matter for the brief you actually want.

Mine are Gmail, Calendar, Slack, Todoist, Drive, Typefully, and Linear. Your list will be different. The right answer is whatever shows up in your daily life.

Each connector walks you through OAuth. Log in, grant access, and pay attention to the permission level. The default for most of these is read-only. Writes, like sending an email or creating a task, usually require explicit approval at runtime.

Keep the first version read-only. Run the brief for a week, watch what it actually does, then decide whether it deserves write access. This thing is going to run unattended on a Mac you are not looking at. Conservative defaults are your friend.

Cowork ships with a Daily Brief template. You can click it, follow the wizard, and have something working in 90 seconds.

Don't do it. It's too generic.

The whole point of running this on your own machine is that you control the prompt. You decide what matters. You decide the format. The preset is fine as a demo, but you can do better in five minutes.

Hit Set up manually, name it Daily Brief, and paste in something like this:

Modify the city, sources, and delivery target. Add a "wins from yesterday" section if you want a little morale boost. Cut the sections you know you'll ignore.

The prompt is a living thing. Plan to tweak it every Friday for the first month. My first version had way too much in it, because of course it did. You'll cut the noise one section at a time.

Do one manual run before you trust the scheduler.

Set the schedule to daily at whatever time you want the brief to land. Mine is 6:30am, which means it's waiting when I unlock my phone.

But before you walk away, hit Run Now.

The first run triggers permission prompts for every connector and tool the agent wants to touch. Read calendar? Approve. Read Slack? Approve. If those fire at 6:30am while you're asleep, the agent stalls on the first one and your brief never shows up.

So run it once now. Tomorrow, it runs on its own with no drama.

One side note: Cowork has fewer delivery options than some other agent systems like OpenClaw. We'll get into that in a future issue.

Run the brief for a week. Each morning when it lands, ask two questions: what did I skip past, and what was missing?

Cut the things you skipped. Add the things you missed.

That's the loop. By the third or fourth version, you'll have a brief tuned to your actual job and life instead of the generic one that ships in every productivity app.

This is the smallest useful agent workflow. It is also the on-ramp. Once you've got the daily brief humming, the same pattern, scheduled task plus connectors plus a custom prompt, is how you build everything else: inventory checks, sales digests, support triage, A/B test reports, the works.

Later we'll get into the bigger workflows that actually justify having a Mac running 24/7, and where Cowork's limits start pushing you toward OpenClaw for certain jobs.

For now, go build your morning brief this weekend.

Share the newsletter