AI employees

Hire the employee you can't find.

Every business has one job nobody wants and nobody can hire for. Chasing invoices. Filing documents. Building the same report every Monday. I install an AI employee that owns that role, and I manage it so it actually gets used.

Scoped after the assessment, and your $999 comes off the top.

Start With the Assessment Talk to one right now

A role, on staff

You are hiring a role. The software is the small part.

A tool waits for you to figure it out. An employee owns work. Every AI employee I install comes with all six of these.

A written job description

One role, defined. What it owns, what it never touches, and what done looks like.

A Company Brain

It knows your services, your pricing, your people, and how you talk. That is why it does not sound like a chatbot.

Only the tools it needs

It only reaches the tools its job needs. Nothing else is reachable.

It asks you first

It asks you first before sending, spending, or deleting anything.

A manager

Me. Monitoring, repairs, and a monthly workflow review. When it breaks, that is my problem.

A weekly value ledger

Every Friday: what it did, the hours returned, and what that time is worth, in numbers you agreed to.

Starting roles

What one request looks like.

Six roles I hire for most. These are scenarios so you can see the shape. The role we actually hire for comes out of your assessment.

Admin desk

One request: find the inspector's email, pull the invoice for that job, make a new job folder, file the PDF, put the numbers in the tracking sheet, send back the link. Minutes later it is done and you never opened your inbox.

Executive assistant

Before you are up: your calendar for the day, the three emails that actually need you, and the follow-up that was about to slip, drafted and waiting on your yes.

Research and prospecting

Fifty target companies checked overnight. The twelve worth your time, scored, each with a first-touch draft in your voice. You approve. It sends nothing on its own.

Recurring reports

The Monday scoreboard builds itself Sunday night. Sales, jobs, cash position, pulled from the tools you already use, on one page you actually read.

File and document management

Every contract, COI, and permit renamed, filed, and findable. Ask for the signed vendor agreement from last spring and get the file, plus the clause you needed.

Content support

Your last client call becomes three posts, a follow-up email, and an FAQ answer. Drafted in your voice, queued for approval, never published without you.

The first time it hands back a finished piece of work you hate doing, you will ask the only question that matters: what else can it do?

Proof

Meet my own team.

I do not sell anything I do not run. This is the staff working in my own business today.

The Radar

Research

Scans roughly 50 companies' feeds every weekday, scores what matters against my criteria, drafts the outreach, and lands a digest on my phone by 8am.

The Producer

Content

Transcribes raw clips, drafts the hooks, renders finished captioned reels. I approve every hook myself. It is the pipeline behind 6M+ views in six months.

The Chief of Staff

Operations

Builds my morning brief at 7:45 every weekday and runs my Telegram. I text it, it delegates to the rest of the team and reports back.

The Bookkeeper

Finance

Takes a folder of bank and card statements and returns a full money report: categories, subscriptions, cancel candidates, flags. Nothing leaves my machine.

My Company Brain

Knowledge

The layer the rest run on: what I sell, who I know, how my projects actually work. It is why their output sounds like me.

The Bottleneck Agent lives on this site. It found the bottleneck in someone's business while you were reading this page.

Talk to it

The knowledge layer

The Company Brain.

Most AI fails in a business for one reason: it knows nothing about the business. The Company Brain fixes that. During the assessment I sit with your team and map how the work actually happens. That becomes a structured knowledge base: your services, your pricing, your processes, your people, the way you talk to customers.

It lives in files that belong to you. You can read every line it knows. Every AI employee you hire runs on it, which is why the second hire trains faster than the first.

The brain is never the product on its own. A knowledge base sitting on broken processes changes nothing. It is built from the assessment and shipped inside the install.

The managed layer

A working agent does not create adoption. Management does.

Most AI projects die after handoff. The build works, the team never touches it. So I stay on as the manager, and this is what that looks like every month.

A three-member group chat

You, the employee, and me, in one channel. Telegram by default. Or Buzz, Block's new workspace built for humans and agents working together, if your team lives in a workspace.

Observable coaching

You watch how I talk to it. I watch what you ask it, catch the unclear instructions, and build the next workflow from what you actually need.

Monitoring and repairs

I watch the system, catch the failures, and fix them. A monthly review keeps the role growing instead of going stale.

A shared source of truth

Important output lands in spreadsheets and documents you can open, check, and edit. Research never disappears into a chat history.

Every Friday, the ledger.

What it did, the hours returned, and what that time is worth, priced at baselines you agreed to. If it does not know how long a task normally takes, it asks you. Estimates you correct, never numbers invented to look good.

Weekly value ledger · illustrative example

Chased 14 unpaid invoices	3.5 hrs	$263
Filed 41 job documents	2 hrs	$150
Built the Monday scoreboard	1.5 hrs	$113
Drafted 9 follow-up replies	2 hrs	$150
One week back	9 hrs	$676

Illustrative week at $75 an hour. Your ledger runs on your numbers.

The part I refuse to wing

AI agent security is what I do all week.

I work in cybersecurity, on AI agent security specifically. My job is learning how these agents break and showing companies how to lock them down. Your AI employee gets the same treatment: it only reaches the tools its job needs, it asks you first before anything consequential, and a human signs off where it counts. An employee you cannot trust with the keys is not an employee.

How it starts

Assessment first. Then the hire.

Find the role.

Start with the $999 assessment. We find the function eating the most hours.

Hire your first AI employee.

Scoped after the assessment. Your $999 comes off the top. One role, fully owned, proven on the ledger before we talk about a second.

Build the full team.

Scoped after the assessment. When the first role proves itself, we use what we learned to scope the next one.

Expand the role before the headcount. The second hire comes when the first one has proven itself.

Straight answers.

What can it actually access?: Only what its role needs. Access is scoped tool by tool, and anything consequential, sending money, deleting records, contacting a customer, waits for human approval. Defining those boundaries is the first thing we do.
What if my team doesn't use it?: Then I have failed, and fixing that is exactly what the monthly fee buys. Most AI projects die after handoff because nobody manages adoption. Your team watches the employee work in the shared channel from day one, and if usage drops, I treat that as a bug in the workflow, and I fix the workflow.
Which role should I start with?: The one you dread. For most owners that is the admin desk or the executive assistant work. The assessment answers this precisely: we find the function eating the most hours and hire for that one first.
What tools does it work with?: Email, calendars, drives, spreadsheets, accounting systems, CRMs, and most software your business already runs. If it has an export or an API, it is probably reachable. We confirm the exact list during the assessment.
Is my data safe?: I work in cybersecurity, on AI agent security specifically. Your employee only reaches the tools its job needs, it asks you first before anything important, and its Company Brain lives in files that belong to you. You can see everything it knows and everything it did.
What does it cost?: I scope it after the assessment, and your $999 comes off the top. When one employee proves itself and you want more of the job covered, we scope that the same way.
Why pay monthly when Claude is $20 a month?: $20 buys a tool that knows nothing about your business. The monthly fee buys someone who hosts it, fixes it when it breaks at 11pm, ships the thing you asked for on Tuesday, and sits with you until you know what to ask it. Nobody is buying the software. They are buying the fact that it keeps working and they never have to think about it. Compare it to one part-time hire who takes vacation and quits.
How do I know it is ready?: Installation is not acceptance. Before you rely on it, I try to break it: missing information, conflicting instructions, a request outside its job, a dead connection, the wrong recipient, and a restart. It passes those on real work without taking live action, and then you see the same test results I do.

One role. Off your plate. Managed.

Start With the Assessment Not sure which role? Ask my agent

Three minutes on your phone. It finds the bottleneck, I find the hire.