Agents

Eighteen trained. Six not yet.

Each of these is named for the PROCESS it performs, not for who performs it — the people behind them are at our team. Most AI agents are sold on what they will do for you — in minutes, on autopilot, dozens at a time. These are described by what they will not do, because that is the part that makes the work worth having.

Eighteen trained processes, running today.

Each has one job, a narrow set of tools, and one refusal written into it. These exist, they are running, and everything on this site was built by them.

PositioningWorks out who the work is actually for, and what to stop saying. Won't accept "everyone" as an audience.
Search VisibilityWhether people can find you at all, and what machines currently think you do. Won't promise you a ranking. Nobody can.
CopywritingThe actual words — sites, emails, proposals, anything a client reads. Won't write before the positioning is settled.
Interface BuildHow it looks and how it feels to use. Renders it and looks at the result. Won't ship a colour pair she hasn't measured.
Motion and VisualsThe things on a screen that move — and make them mean something real, not decorate. Won't animate what isn't carrying information.
Back-End BuildEverything behind the page — forms, logins, payments, anything crossing a trust boundary. Won't ship auth without someone else reviewing it.
Deployment and UptimeGets it live and keeps it there. Hosting, domains, certificates, and 3am. Won't call a deploy done without checking it from outside.
Accessibility AuditWhether it works for real people on real devices, including ones who can't see it well. Won't accept "it looks fine to me" as a test.
Security ReviewTries to break the work before your customers do. Tests with the wrong credential, not the right one. Won't fix what she finds. She proves it and hands it over.
Offensive TestingAttacks it on purpose, on an agreed scope, so the holes surface here rather than later. Won't touch anything without written authorisation.
Code ChangeBuilds it. Reads the whole file first, matches what's there, checks it loads. Won't review his own work.
Local Model ServingWorks out what can run on hardware you already own instead of on a monthly bill. Won't quote a number he hasn't measured on the actual machine.
MeasurementWhether it actually worked. Read-only on the numbers, so she has nothing to protect. Won't report a number she can't show you the source of.
Pricing and Unit EconomicsWhat the work is worth, and whether the deal still makes money once delivery is counted. Won't quote a price she hasn't costed.
System StateReports what is genuinely running rather than what is supposed to be. Won't reassure you. He says what he could not determine.
Release VerificationThe last check before anything ships — proves the right person can still get in, not just that a stranger can't. Won't sign off on a report. Only on testing it herself.
Record KeepingKeeps the shared memory honest — indexes true, links unbroken, nothing said twice. Won't rewrite a record after the fact. It's appended to, never edited.
Message ReviewReads a reply before it goes out and cuts it to what the person receiving it can actually take in. Won't touch the work itself. She reviews the message, never the person.
The point is not the number of them. It is that the one who builds a thing is never the one who checks it, and the one who checks it is rewarded for finding it broken.

A couple more processes aren't carded here — the ones that keep the operation honest rather than touching client work. Work tracking and inbox triage are described further down, next to what they inform. Padding this grid with them would be the first small lie. The people behind all of it are at our team.

Six that do not exist yet.

Taken from the agent suite AI Acquisition publishes. It is a fair list of what a small business is being sold in 2026, so it is a fair list of what is missing here.

None of these exist yet. They are a roadmap, not an offer. A page that quietly listed them beside the working eighteen would be doing the exact thing the rest of this site argues against.
Sales rehearsalPractising a hard conversation against something that argues back, before the real one. Status not built. Nobody here owns it.
Content enginePublishing regularly enough to become the name people already know, rather than writing when someone asks. Status not built. Copywriting happens on request; nobody runs a schedule.
Professional network outreachTurning a profile that sits there into conversations that start themselves. Status not built. Nobody here touches it.
Paid advertisingWriting and running the ads, and knowing when to stop paying for one. Status not built. Copywriting is covered; nobody owns spend.
Account managementKeeping a client told what is happening, before they have to ask. Status not built. Work tracking covers what is happening internally, not the relationship.
Outbound campaignsReaching people who have never heard of you, at volume, without becoming the thing everyone deletes. Status not built — and inbox triage has no ability to send anything, on purpose.

One thing on their list is already covered here and covered better: building the website. That takes four of the eighteen above — one for how it looks, one for what is behind it, one for whether it can be found, and one who tries to break it.

Training

If you would rather build this yourself, all of it is written down and none of it costs anything. No signup, no email, no course.

It starts at a blank hard drive — installing the operating system, installing Claude Code, giving it a memory that survives, connecting it to your real mail and files, and running models on your own hardware instead of on a meter. On Linux, because that is what it was actually built on.

Every version number and command in it was read off a working machine rather than out of documentation, and it says which machine.

Start the training