Sorry, no video this week.

I'm trying something a little different. A short written note about something I've seen, tested or learned while running the business.

This week, it's about the difference between AI that can write and AI that can actually find the work and get it done.

A few weeks ago, I had several things bouncing around in my head about invoicing.

There was a speaking engagement that needed an invoice. Two client emails I had to follow up. A website project where I thought we might have missed the deposit invoice. And a group of WordPress hosting clients who had not had a price rise in about five years.

That last one is what happens when your CFO leaves and you don't replace her! (Don’t worry I’m building a new one)

I told my AI agent I would organise all this information and give it a proper brief later.

It stopped me.

"Just give me the raw dump. Any client names, amounts or details you can remember."

So I did. It wasn't particularly structured. It was roughly the sort of thing you might dictate into your phone while trying not to forget something else.

Within about ten minutes, the agent had:

• prepared the speaking invoice in Xero

• drafted replies to the two client emails

• found the website project

• found the original quote and our meeting notes

• identified a $2,500 deposit invoice that should have gone out three weeks earlier

• found the ClickUp task covering the hosting price review

• pulled together the affected hosting clients

I still had to review and approve the consequential actions. I'm not giving an AI unrestricted permission to send invoices and emails on my behalf. I've made that mistake before.

But the work of finding everything, matching it to the right clients and preparing the next actions had been done.

The interesting part wasn't the writing.

ChatGPT, Claude or Gemini can all draft a perfectly reasonable email. That's useful, but it's not difficult anymore.

The useful part was that the agent knew where the work lived.

It could connect a half-remembered comment to an email, a quote, a meeting note, a ClickUp task and the correct record in Xero. I didn't have to open five systems and reconstruct the history myself.

That context gathering is where a surprising amount of business time disappears.

You remember that something needs doing, but not where the conversation happened. You find the email, then discover there was a meeting. The meeting points to a quote. The quote points to a project. Twenty minutes later, you're still assembling the facts needed to start the actual job.

A normal chatbot sits outside that workflow. It only knows what you paste into the conversation. It might write beautifully, but it has no idea what has already happened inside your business.

An agent can be different.

Give it access to the right information, teach it how your business works, and define what it may prepare versus what requires approval. It can then do more than generate another nice paragraph.

It can find the work.

That doesn't mean handing your entire business to an AI and hoping for the best. That would be reckless.

Start with one boring workflow that repeatedly slows you down. Meeting preparation. Invoice follow-up. Turning emails into tasks. Finding the latest client decision. Preparing a report from information you already have.

Let the agent gather the context and prepare the next action. Keep a person responsible for the judgement and approval.

Most businesses don't need more AI-generated content.

They need fewer things bouncing around in someone's head.

What's one boring job in your business that repeatedly requires you to search through emails, tasks, documents and notes before you can actually do it?

Reply and tell me. I'm looking for real workflows to pull apart in future newsletters, and I may show you how I'd approach yours.

Sorry about the lack of video. Normal service may or may not resume next week.