Build & Defend Newsletter 20

One email a week - something from which I hope you'll get real value. We talk about things we can build, and how to defend them. That can apply to cybersecurity, physical buildings, digital products, and .... just about anything. It gives me a lot of latitude in what I can write about, but the two concepts are important for progress - as individuals, and as society.

Today's topic is: your first employee.

For a long time, one of the first hires in a small business was obvious.

An assistant.

Someone to answer emails, schedule meetings, chase invoices, summarize calls, organize files, remind the owner what they forgot, and generally keep the machine from falling apart. In fact, I know to this day a billionaire who still works with his original assistant because they're so in synch that he's WAY more effective than he might otherwise be.

Nowadays, however, she would be an AI instead of a she.

That's right, not a person.

An always-on AI agent.

Not “AI” as in a chatbot you open when you remember to use it, or a desktop app that strangely requires multiple updates when one of their "we-called-it-dangerous" models is determined to be dangerous and then that company is surprised that it has to be removed.

Your always on AI agent sits quietly in the background, watches the right inputs, summarizes what matters, drafts what needs a response, reminds you before things break, and gives you back the one thing every solo business owner is short on:

Attention. And Time.

Sorry, that's two things.

This matters because most small businesses do not fail because the owner lacks ambition.

They fail because the owner becomes the bottleneck.

The inbox piles up. The CRM gets stale. Follow-ups slip. Meetings happen but action items disappear. Bills, renewals, leads, reviews, website errors, and customer questions all compete for the same overloaded brain.

A good assistant helps solve that.

But an assistant costs real money, needs management, needs training, and only works certain hours.

An always-on agent is not a perfect replacement for a human assistant. But it may become the first operational layer many solo founders install before they hire anyone.

And for a lot of businesses, that may be enough to change the game.

There's talk of what will become the first single-person (and multi-agent) billion dollar business. My guess is we'll see that in less than two years.

The practical version of this is not science fiction. Or any kind of fiction. You can do this today. I'm helping real customers solve real world problems with workerbee.bot. (Interested? Check out the site.)

You do not need to build a full autonomous employee on day one. Or ten. Maybe by day .... 366?

Start with a narrow agent. Give it one job. Then another. Then another.

Here are a few easy places to begin.

1. Inbox triage agent

Most people do not need AI to “write all their emails.” In fact, that might actually be annoying for the recipients, when they start seeing all the emdashes. ("--") (No, that's not an emoji.)

Most folks need help knowing what actually matters.

Create a simple inbox workflow:

You can do this with tools like ChatGPT, Gmail filters, Zapier, Make, or more advanced agent setups if you are technical. Of course, you've already built agents based on previous newsletters, right? 🙂

Don't fall into the “let AI run your inbox” trap. Mailmate starts giving you warnings about synch errors when you do.

Instead, "Every morning, tell me what needs my attention and draft the easy replies.”

That alone can save hours per week.

2. Meeting summary agent

If you take calls with clients, vendors, or partners, every meeting should produce a useful summary.

Not a transcript, but a decision record.

After every call, your AI should give you:

This is one of the easiest AI wins for small businesses because the output is immediately useful If you use Zoom, Google Meet, Teams, Fireflies, Fathom, Granola, or Otter, you can already do most of this; for things like Zoom and Teams they will be begging you to do it at the start of every meeting. Literally begging you. Every meeting. With multiple popups in some cases. Not that it's annoying or anything....

The tactical improvement is to create a standard format.

Meeting Summary:

3. Follow-up agent

Most money is lost in the follow-up:

The lead who asked for pricing. The client who said “Let's circle back next month...." who actually meant it. (There are some, though few; I promise.)

The partner who promised an intro.

The prospect who opened your proposal but never replied.

An always-on agent can watch for these.

Even a simple version can help:

This is where AI becomes less like a search box and more like a junior operator.

Not making strategy. Just preventing dropped balls. (I said "balls"....)

4. System watcher

If your business depends on a website, payment page, booking form, newsletter, or online store, you should know when something breaks.

An agent can watch:

This does not need to be fancy. A simple monitor that says “your checkout page is down” or “you received a 1-star review” is incredibly valuable.

You don't need another dashboard, you just need to know when something needs to be fixed.

5. Daily owner briefing

This may be the highest-leverage habit.

Every morning, have your agent produce a short briefing.

Example:

Good morning! Today's weather in [YOUR_LOCATION_HERE] will be terrible. It'll be a great day to be stuck at your desk getting through the drudgery!

Today’s priorities:

1. Reply to Sarah about the proposal.
2. Review the unpaid invoice from Client B. (They really did want to follow up from last month!)
3. Prepare for the 2pm sales call.
4. Website form submissions are down 40% this week.
5. You have 3 follow-ups older than 5 days. It's Friday. No one will read them tomorrow.

Suggested first action:

Send the proposal follow-up to Sarah. You did it, right?

That is management. Admittedly, with snark, but then you couldn't get that from a human and still want to employ them regularly.

Your agent needs to be able to remember things. This is kinda what agents do, to an extent, but they potentially need to be able to connect to your business information as well. Give them read only access to your Google Docs/Notion/SQL/APIs/FileShares/etc.

Your agent shouldn't write for you right away. When it does, it should use the Hermes "humanizer" skill. (OpenClaw has something similar but the Hermes one is better.) But start with things as read-only until you're to the point where you trust it.

Do not give your first agent full permission to send emails, delete files, move money, or change systems. You can let it read your inbox, draft replies (in its chat window, perhaps at first, then later in the "Drafts" section) and you need to approve the emails before sending. It should summarise meetings and create reminders. It can suggest actions, but not take them.

Wait, doesn't that defeat the purpose of an agent?

Well, kinda, yes. But then again they can be hallucinatory and live in a tiny box on your desk. If your first human employee hallucinated sometimes and lived in a box on your desk you wouldn't trust them right away, would you? Or at all? But I think you get my point. You can have it do the actions but after you've vetted it.

We talked about permission surfaces in Newsletter 18. You want the agent to have the least permissions possible, and to only do things you've vetted over time. It only takes one email to a customer that says "y0u S4ID Y0U'd f00lloW UP L4st M0nt|-| W33k(EnD)|" for your customer to really worry about continuing to do business with you.

Require approval for:

Overtime, you can let the agent act automatically on low-risk work:

Then give it bigger tasks.

A good small-business agent does not need full autonomy. It needs good judgment about when to stop and ask.

The other thing - maintain an audit trail. The benefits of this should be obvious, but if you need to, you can go back and figure out what the agent did. Remember when I thought that Talos was building me things but he was just telling me he was and not actually doing the things? Yeah, that sucked, but because I had a record of our conversation, I could go back and figure out what happened.

Perhaps you haven't yet built an agent. I can help with that (see above (or below)). If you build one. yourself, then you can give it a habit stack:

General things like that will get you a long way with very little effort.

WorkerBee.bot
If you'd like me to help you build your first agent(s).
WorkerBee.bot

Zapier AI Automations  
If you don't go the agent route right away, it's good for non-technical workflows across Gmail, Sheets, Slack, Notion, and thousands of apps.
https://zapier.com/ai

Make  
Visual automation builder. It can be more flexible than Zapier once workflows get more complex.
https://www.make.com

n8n
Open-source automation platform. Good for people who want more control or self-hosting.
https://n8n.io

Fathom
AI meeting notes and summaries. Useful for sales calls, client calls, and internal meetings.
https://fathom.video

Fireflies
Meeting transcription and summaries across common meeting platforms.
https://fireflies.ai

UptimeRobot
Simple website and endpoint monitoring. Great first step for system watching. (I've been a customer for a while.)(They didn't pay me to say that, though that would have been nice.)
https://uptimerobot.com

Uptime Kuma
Open-source self-hosted monitoring tool.  
https://github.com/louislam/uptime-kuma

Tailscale  
Private networking for securely connecting your devices and servers. Useful if you self-host. Which you will. Eventually.
https://tailscale.com

Pick one repetitive business task this week and turn it into an agent-assisted workflow.

Thanks for reading this newsletter! Feel free to respond any time.

Thomas

Was this forwarded to you? Subscribe at builddefend.fyi.

Had enough? [Unsubscribe] - no hard feelings.*

* Well, a little bit.