5 things my AI agent actually does every day

I see a lot of “100 things you can do with AI agents” posts and they’re always hypothetical. Theoretical use cases that sound cool but nobody’s actually doing them. “Have your agent negotiate with vendors!” Sure thing buddy.

Here’s what mine actually does. Every day. Right now. On a Mac Mini sitting on my desk.

1. Checks my email and tells me what matters

I get probably 80 emails a day and maybe 5 of them matter. My agent scans my inbox a few times during the day and sends me a summary on Telegram. Important stuff gets flagged. Newsletters and promotional stuff gets ignored. I haven’t opened Gmail in weeks.

The key here is that I didn’t build some complex email classifier. The agent just reads the subjects and senders, applies common sense, and tells me what looks urgent. LLMs are surprisingly good at this.

2. Monitors my calendar

Every morning (and a few hours before any event) the agent checks my calendar and gives me a heads up. “You have a call in 2 hours.” Simple. But I used to miss stuff all the time because I’d forget to check my calendar. Now I don’t.

3. Runs background research

When I’m working on something new, I tell the agent to research it. It searches the web, reads articles, summarizes findings, and drops a report in my workspace. I come back 20 minutes later and there’s a document waiting for me instead of 30 browser tabs.

This is probably the highest-value task. The amount of time I used to spend reading through search results and articles is wild. Now I just read the summary and dig deeper only when I need to.

4. Manages files and notes

I work with a lot of markdown files. Project docs, notes, plans, whatever. The agent can create, edit, search, and organize all of it. “Add this to the project plan.” “What did we decide about the pricing?” “Create a new doc for this feature.”

It’s like having a shared workspace with someone who has perfect memory of everything in it. Because it literally does. It can search through all my files instantly.

5. Keeps me honest about tasks

I have a list of things I want to get done each week. The agent checks in during its heartbeat cycle and if something’s been sitting untouched for too long, it’ll nudge me. Not in an annoying way (I configured that). Just a “hey, you haven’t touched X in 3 days, still want to do it?”

It’s accountability without another human. Which sounds sad but it’s actually great because I don’t have to explain to anyone why I procrastinated. The agent doesn’t judge. It just reminds.

What I DON’T use it for

Anything that requires nuance in communication. I don’t let it write emails to clients. I don’t let it post on social media without me reviewing it first. I don’t let it make purchasing decisions.

The pattern is: information processing and reminders, yes. Representing me to other humans, no. At least not yet.

The setup

I use OpenClaw for all of this. It runs locally on my machine, connects to Telegram so I can talk to it from my phone, and uses Claude as the underlying model. The whole setup took maybe 15 minutes.

If you want to try something similar, I wrote a setup guide and I have some free templates that give you a pre-configured starting point.