The AI Edge

The Briefing

Klaviyo turned its K:AI into a full marketing agent this month, and HubSpot pushed its Breeze agents further. These do not just suggest. They build whole email journeys and close support tickets on their own. The old idea of a "workflow" you set once is starting to look old. The machine is moving from helper to doer.

Send timing just changed under our feet. Apple Mail Privacy hides open data for about half of all inboxes, so opens are now a lie. The smart platforms have moved to clicks and real actions to guess the right moment. If your timing still runs on open rates, you are aiming at a ghost.

Meta rolled out brand-aware creative generation inside Ads Manager. You feed it your look and your tone, and it drafts ads that sound like you, not like a robot. Meta also added AI connectors so you can run your ads from tools you already use. Less clicking around. More making.

Google launched Ask Advisor at Google Marketing Live. It is a Gemini assistant that sits across Ads, Analytics, and Merchant Center, and you just ask it questions in plain words. One place. One conversation. The whole stack answers back. This is the direction every platform is now walking.

The Build

This week I did not build an automation. I killed several.

For years I ran the same play everyone runs. Someone joins the list, a timer starts, and the same five emails fire on the same schedule for every human on Earth. It works. It also feels like a vending machine. And I've started to hate the taste.

So I spent the week on something harder. I wanted to teach AI to write one intimate message, to one person, and to tell me the right moment to send it. Not a blast. A knock on the door at the hour that person actually wants to hear from you.

Here is what I believe now.

As AI gets stronger, the old CRM automation dies. It gets replaced by real AI interactions.

The beast serving the human.

And that is what people have wanted from AI all along: a deeper, more personal, more timely connection to the source, the kind that buys back time, money, and freedom. The machines can do it. But you have to train them differently.

And no, it's not one md.file.

I started with the CRM framework I use with clients. The rule there is simple: pull every signal about a person into one place before you write a word. What they bought. What they clicked. What they went quiet on. When they last showed up. Most people skip this and wonder why their "personalised" email feels like a form letter.

Then I gave the machine a real job. Take one contact. Read their signals. Write to them like I would if I remembered every detail about them. And tell me when to hit send.

This was the prompt:

You are my writing partner. You know my voice cold: warm,
plain, one idea, short lines, no hype.

Here is ONE person from my CRM:
- Name: [name]
- Last bought: [product], [date]
- Last 3 clicks: [links/topics]
- Went quiet after: [event/date]
- Notes I wrote about them: [personal details]

Do three things:
1. Tell me what this person is probably feeling right now,
   based only on the signals above.
2. Write ONE short message to them in my voice. Make it feel
   like I remembered them, not like a campaign.
3. Tell me the single best day and hour to send it, and why,
   using their click and action pattern (not open rates).

If a signal is missing, say so. Do not invent it.

I ran it against 12 real contacts. Then I read every draft out loud, the way I read everything before it goes out.

What worked: The timing calls were sharp. On 9 of 12 people it picked a send window I would not have guessed, and it showed its work from their click pattern. The writing had real warmth when the notes were rich. One message about a client who went quiet after a hard month made me stop, because it sounded like me on a good day. That is the bar.

What didn't work: When the notes were thin, the writing went generic fast. Three drafts read like every other email in the world, because I had given it nothing human to hold. The machine can only feed me machine learning. It cannot feed me human learnings. Those still come from me knowing my people. It also over-trusted one signal twice and I had to pull it back.

The lesson landed hard. Each time I think I am close, a new layer shows up, and I have to keep going on my own path of learning. AI does not hand you the answer. It hands you a faster draft of your own thinking, and only if your thinking was worth something first.

The Play

If you are someone who wants to truly connect with your audience, here is the move. Stop writing to your list. Start writing to one person on it.

Pick your five best clients or subscribers. Not a segment. Five real names.

Write down everything you know about each one in plain words.

Then run a version of the prompt above, one person at a time. Read each draft out loud. If it does not sound like you talking to a friend, the notes were too thin, so go add more.

Do this for a month and something shifts.

You stop thinking in campaigns and start thinking in people. The AI does the heavy lifting on timing and first drafts. You do the part no machine can do, which is actually caring what happens to the person on the other end.

Before any of this works, your data has to live in one place. That is the boring foundation, and it is the whole game. The CRM and customer data framework below walks through how to pull every signal together so your writing partner has something real to work with.

Try this today. Open your CRM, pick your single best client, write three true things about them that are not in any database, and use the prompt above to draft one message plus the right time to send it.

Then read it out loud before you touch send.

Download crm-customer-data-1.md