What my AI agents actually do all day

What my AI agents actually do all day

Most posts about AI agents are hypothetical. "Imagine if an agent could…" followed by a demo that works once and breaks in production.

I got tired of imagining. So I built three agents that run my business and my family admin. They've been live for weeks. Here's what a normal day looks like.

6:45am. Coffee. News briefing I didn't write.

Every morning, Hermes sends me an AI news briefing. Everything notable from the last 24 hours, waiting in my inbox before I've opened my laptop.

I used to spend 30 minutes scanning Twitter, newsletters, and Reddit to piece this together myself. Now it's done. The setup took less time than writing this paragraph. Hermes connects to my Gmail, so the briefing just lands like any other email.

I skim it over coffee. Some mornings it surfaces something I'd have missed entirely. Some mornings it confirms nothing happened. Both are useful.

Every three days. The business check-in I never have to chase.

This one changed how my co-founder and I communicate.

Every three days, Hermes emails both of us asking for updates. Not a vague "how's it going?" It's structured.

It covers our open tasks and their current status. Our pipeline of prospects. Current customers, their value, how much we've invoiced, any notes. Proposals we've sent out and where the chase is at. Active leads by owner. Lost deals. Then it asks us specific questions to reply to.

We reply on email. Hermes processes those replies and keeps our prospect data up to date.

Before this, the updates lived in our heads, a spreadsheet, and a WhatsApp thread. Now there's a single source of truth that rebuilds itself every 72 hours. The agent doesn't just ask the questions. It listens to the answers and acts on them.

Friday afternoon. The school newsletters I never read properly.

This one's personal, and honestly it's the one that gets the best reaction when I tell people about it.

Our kids are at two different schools and a nursery. Each one sends newsletters on Fridays. Attachments, PDFs, links to forms. The kind of thing you mean to read properly but end up skimming at 10pm while half-watching TV.

I set up a separate agent to wait until Friday afternoon, once all the newsletters have landed. It reads every newsletter and every attachment. Then it sends me and my wife a summary of actions we need to think about for next week. After-school clubs to sign up for. Events coming up. Forms that need filling in. Permission slips with deadlines.

We went from "did you see the email about the trip?" on Sunday night to having a clear list by Friday evening. Last term it caught a reception show that I would have missed entirely. The kind of thing that, if you'd missed it, you'd be hearing about for years.

Saturday morning. The weekend coaching note for our son.

This is the one I didn't expect to get the most out of.

The same agent that handles the newsletters runs a second cron job on Saturday mornings. It tracks our son's progress in maths, comprehension and phonics, and emails us a short note: how he's doing, and three things to try with him this weekend at his current level.

We do a few pages of a workbook with him, I take a photo, and send it back to the agent. That photo becomes the input for next week's note. The agent recalibrates what level he's working at, what he's struggling with, and what to nudge next.

It's not teaching him. We do that. The agent coaches us on what to try, which is the bit that's hardest to do consistently when you're tired and trying to remember what last weekend's worksheet looked like. It's the only reason I have an honest, week-on-week sense of where he is.

How I set this up

The tool is Hermes. I'd been curious about AI agents for a while, but what surprised me was how little friction there was.

I connected it to my Gmail, calendar, and Google Drive. I set up voice input so I can drop voice notes instead of typing instructions. I set up email as both an input and output channel, so the agents can send and receive.

That's it. No code. No API keys pasted into terminal windows. The agents I described above were configured in plain English.

I've also used Hermes to set up meetings with our team ahead of calls. I describe who's attending and what we need to cover, and it handles the calendar invite and agenda.

What's next

I'm extending this. Pre-meeting briefings, so I walk into every call with context on the person, their company, and what we last discussed. Scheduled SEO audits for Dreams AI Can Buy. Automated tracking of how our content performs across social channels and the website.

The pattern is always the same: find something I do repeatedly, describe it to the agent, and let it run.

The honest bit

Some things needed adjusting. I tweaked the business update prompts twice to get the right level of detail. The newsletter agent occasionally misses an attachment if the school sends it as a separate follow-up email.

But the baseline is solid. These AI agents save me real hours every week. Not theoretical hours. Real ones I can point to.

If you're curious about what this kind of setup looks like in practice, I run 90-minute AI training sessions where we build things like this live — see what happens in a session. For day-to-day file and knowledge work, I also run Claude Cowork training so the same agent that drafts your email can read and write the files that matter. And if your business wants to go deeper with agentic workflows, Exponential Partners is where that happens.

Or just message me. I'll tell you honestly whether it's worth your time.