I Sell AI Brain Installs. I Scored 10 out of 18 on My Own Diagnostic.

I charge £6,500 for an AI Brain install. This week I ran the client diagnostic on my own business and scored 10 out of 18.

It’s the same 6-category instrument I run at the start of every engagement. The site has mentioned it for months without ever naming the categories. Here they are, with my scores. Each scores 0 to 3.

Three of those categories are also the three pillars of a Brain: Context, Workflows, Archetypes. The pillars are the architecture. The diagnostic measures whether the architecture is actually running. Mine mostly isn’t.

One more detail, and it’s the one that stung. My own self-install, the exact process I sell, has been stalled at step 1 for 10 weeks. Step 1 is answering questions about my own business. It isn’t a technical blocker; it’s a founder who wouldn’t sit still long enough to be interviewed by his own product.

Connected is not the same as running

Here are the receipts, since a score without them is just a confession.

I have 11 connected data stores, including Xero, HubSpot, Gmail, Notion, and Linear. On paper that looks like a finished install. Connection is the part that demos well.

And yet I counted more than 50 manual context pastes into Claude in a single week. My top three paste sources are Slack, email, and Google Drive. Two of the three are already connected: Gmail and Drive. I paste email threads into a chat that can already read my email. The connectors exist. I still paste.

The automation tells the same story, just more politely. Eleven scheduled tasks run on their own: research digests, a competitor watchlist, periodic checks. That’s why Workflows scores a 2 and not a 1. But notice what they have in common: they all watch, and none of them does. The agents that touch the core of the business tell the other half. There’s a sales agent that drafts follow-ups off the booking form, and a 5-agent content pipeline that produced the post you’re reading. They’re all real, tested, and every one is a script I trigger when I remember to. They have no role, no schedule, and no escalation rules. So they’re scripts, not colleagues.

When I asked myself why the score was low, I wrote this down, verbatim: “not having enough time to ask AI to do everything for me. I don’t have enough screens.”

Read that back. The constraint I named was screens. It wasn’t model quality, and it wasn’t tooling or budget. It was screens. My AI capacity is capped by my attention, not my subscription. And the honest name for it is this: nothing in my business runs unless I press the button.

I am the dispatcher

That’s the whole diagnosis, and it’s why the score is 10 and not 15.

I get roughly 20 hours a week back from AI today, most of it through Claude Code. That number sounds like a flex. It’s evidence of the ceiling. All 20 hours are pull. I ask, it answers, I ship. Nothing is push. Nothing arrives without me initiating it. The return scales with my attention, and my attention was fully spent before AI turned up.

If you live in Claude or ChatGPT daily, you know the shape of a pull day. You open the tab, you feed it context, you get good output, you close the tab, and the business goes quiet again. The quality of the output climbs every quarter. The shape never changes.

My bet is 40 hours a week after a full install. The second 20 cannot come from me asking better questions faster. It has to come from work that happens while I’m not looking. That’s the gap between connecting AI to your business and AI running parts of your business, and no model upgrade closes it.

It’s the gap the Archetypes row measures. An archetype is a written job description for an AI colleague your business needs but hasn’t hired: a role, escalation rules, a quality bar. Archetypes are the most durable layer of a Brain because they describe roles in your business, not tools. They outlast every model underneath them. I scored 0. I had role definitions in draft and nobody on shift.

One named role found work nobody assigned

Same day as the diagnostic, I installed the first archetype: an Operations Manager whose job is a weekly operating review that runs whether or not I show up.

I dispatched the first run myself, as proof the install works, and the unattended schedule starts Monday. Even hand-triggered, it surfaced overdue invoices and stale deals I had stopped seeing. There was nothing exotic and no clever prompting. I pressed the button, but I didn’t choose what it looked for. The job description did. A named role went and did its job.

I could have found every one of those items myself in 20 minutes. I hadn’t, for weeks, because dispatchers don’t do reviews. The value of the archetype isn’t intelligence. It’s that the looking no longer waits on my attention. From Monday, neither does the trigger.

One archetype does not turn a 10 into a 14, and I’m not claiming it has. The score moves when the review has run unattended for weeks, when the content pipeline has an owner, when the watchlist escalates instead of waiting to be read. It’s measured, not claimed. I’ll re-run the diagnostic and publish the number either way. A system that needs its installer is a failed installation, and right now I’m still the installer of almost everything I run.

Score yourself. Two minutes.

It’s six categories, one question each. Score each 0 to 3: 0, no. 1, it exists but only runs when you drive it. 2, it runs, but you’re still in the loop somewhere. 3, reliably yes, without you. That middle band is where most of my own scores sit, and probably yours.

• Context. Can your AI answer a real question about your business, pipeline, cash, customers, without you pasting anything in?
• Workflows. Do your repeatable processes, proposals, reporting, follow-ups, exist as saved, named workflows rather than prompts you reconstruct each time?
• Archetypes. Is there any named AI role in your business that does its job on a schedule without being asked?
• Voice. Does AI output ship in your voice without a rewrite pass?
• Team activation. Can anyone else in the business, human or agent, use the system without you brokering it?
• ROI clarity. Can you state what the system returns each week, in hours or pounds, and would the number survive an audit by someone who doesn’t like you?

If you scored 10 or under, you’re where I am: a dispatcher with excellent tools. You’re proud of the setup and capped by your own attention. The cap doesn’t move with a bigger subscription or a better model. It moves when roles run without you.

I don’t have everything landed yet. That’s not a confession, it’s the process: an install is work you do, not a state you declare, and I’m doing mine in the open with the scores published as they move. It’s also why the first cohort runs at a reduced pilot rate. Ten founders install alongside me while I finish my own, and we debug the process on both of us.

If you’d rather run yours alongside mine, that’s what the course is: 6 weeks. Week 1 is this exact diagnostic, scored honestly. Week 6 re-runs it. If the number didn’t move, the work didn’t work. Join the waitlist.