I was managing accounts before there was a button for any of this.
Last week, before I touched a single campaign in a new account, my agent did something I would have done myself. It opened a document I wrote called my Google Ads management principles, read it top to bottom, and only then started the audit. Twenty years of “don’t echo Google’s recommendation,” “don’t bulk-apply a bid change across keywords that carry their own bid,” “leave the paused campaigns alone, because that history is worth more than the tidiness.” It read all of it before it looked at a single number.
That document is twenty years old. Not the file. The judgment in it.
And it does that because, twenty years in paid search since 2005, I was gluing my own tools into my work long before any of this had a toggle. Not because I’m clever. Because I had to. Most of those two decades was spent connecting things that weren’t supposed to talk: spreadsheets to scripts, scripts to ad platforms, my own head to a filing system I could half-remember. So when the agent era arrived and you could suddenly hand it your email, your notes, your drive, I did the obvious thing fast. I pointed it at the folder where all my client docs live and asked it what it made of one of the accounts. I gave it a meeting recording and asked it to tell me what we agreed. I watched it answer from the real thing instead of a copy-paste.
And it was genuinely good. For about three weeks.
Then it plateaued. The connectors did exactly what they promised, which turned out to be the ceiling, not the floor I’d hoped for. The agent could read my inbox, but it triaged it like a stranger would, with no idea which sender is a client mid-renewal and which is a newsletter I keep meaning to cancel. It could read my meeting transcripts, but it summarised them the way anyone would summarise anyone’s meeting. It had access to everything, and judgment about nothing. The plumbing was on, but the water came out lukewarm. It just wasn’t me.
What I actually built sits behind the obvious connectors, not in them.
The hero of it isn’t a connector. It’s a set of what I call skills, which is a slightly grand word for “the way I do a specific job, written down once so I never have to re-explain it.” I have one called process-receipts that handles finance mail end to end, because I got tired of telling the agent, every single time, which vendor goes where and which payment-failure needs me and which renewal is fine to ignore. I have meeting-ingest, which takes a fresh recording and files it the way I’d file it, into the buckets I actually use, with the action items pulled the way I pull them. There’s inbox-sweep, which triages mail against rules I wrote, not against generic-helpful-assistant instinct. There’s vault-sync, which writes what we learned in a session back into my notes so next week starts ahead of where this week ended. And process-content-card, which turns a rough idea on a board into a draft in my voice, not in the flat house style of a model that’s never met me.
Underneath all of those sits the thing I built first, the one I’ve written about before: a second brain in Obsidian. A plain folder of plain text files that holds how I think about clients, about offers, about the work. The skills are how the agent does a job. The second brain is what it knows before it starts. The connectors are just the pipes that carry the water between them.
That management-principles document is the piece I’m proudest of, precisely because it’s the most boring. It’s not a clever prompt. It’s a rulebook I keep adding to, sitting between the connector and the account, so the agent works like the most senior version of me on a tired day instead of like a confident graduate who read the help docs once.
There’s even a face on the client side. His name is Bert, he’s the AI persona that handles some of the client-facing back-and-forth, and the clients know exactly what he is. We introduced him to a client group, by name, as an AI. Nobody’s being fooled. He works because of everything behind him, the skills and the second brain and the principles, not because he’s pretending to be a person. He is the second brain wearing a name.
None of this was a weekend.
I want to be honest about that, because the toggle-it-on framing makes it sound like an afternoon, and it wasn’t. Every skill I have exists because something broke first. process-receipts came from getting it wrong by hand. The principles doc grew one line at a time, usually the day after I caught the agent about to do something I’d never do. I didn’t sit down and architect a system. I wired things after the pain, not ahead of it, one annoyance at a time, and most of what I built started as me muttering “I am never explaining this again” into a chat window. It was slow. It was unglamorous. It compounds, but only because I kept showing up to it.
So this is the last note in a four-week run, and the honest closer is also the simplest one. The connectors are the floor. Turning them on is the easy part, and everyone will have done it by next quarter, which means it stops being an edge the moment it’s a button. The building is yours. The skills that hold how you work, the notes that hold what you know, the rulebook that carries the judgment you spent years earning, that’s the part nobody can ship as a feature, because it’s made of you. The second brain, the plumbing, the twenty years, they’re one thing, not three.
And it’s worth nothing without that judgment. An extension is never better than the one it extends. Take a sharp operator and give them this, and you get something close to frightening. Take someone with no judgment, give them the exact same setup, and you get confident nonsense, faster. You can’t download my second brain. You’d have to live my twenty years to fill it.
So don’t. Go live yours, and build your own.