Your First AI Employee Is A File You Haven't Written Yet

What You’ll Gain

• The three layers of an AI employee: the seat, the jobs and the procedures, and which one actually matters.
• The difference between a procedure and a workflow, and why a business is built out of both.
• Why an AI employee can be more consistent than a person: it rereads its whole job description every time it runs.
• Where your AI employees should live, and the problem with renting them one subscription at a time.
• How to ask your business numbers a question by connecting your AI to the tools you already use.

Biggest Takeaway to Implement: Write down one procedure in your business, step by step, in enough detail that an AI could follow it without asking you a question. That is the layer nobody wants to write, and in Chris’s words it’s also the only layer that matters.

0:00 The Biggest AI Shift Of The Next Year Isn’t A Model

Chris hosts solo while Olga is away. He sets up the episode around what he thinks is the biggest AI topic of the next year: AI employees, or as he prefers to call them, AI operators for your business. The plan: a quick run through the news, a long look at what an AI employee is and how it works, then funding.

1:32 Half The Country Uses AI. Only 18% Trust It.

NBC News polled 7,105 adults. 52% now use AI often or sometimes, up eight points from June last year. 70% are more worried than excited. Only 18% trust what it tells them most or almost all of the time: 2% almost always and 16% most of the time. The most common answer was “only some of the time,” at 47%. Chris’s read is that there are two groups. Casual users drift with the sentiment. Serious users are not looking to AI for truth. They use it to do procedural work they would otherwise do themselves, and they still check what it produces.

5:09 Ask Your Business Numbers A Question In Plain English

OpenAI shipped a data agent: connect ChatGPT to wherever your business keeps its numbers and ask questions like “which customers haven’t ordered since June?” It writes the query, runs it and builds the dashboard. It is a limited rollout, on ChatGPT Work and Codex plans with an administrator switching it on. Chris’s bigger point: you do not need to wait for this feature. Connect whichever AI you use to the APIs or MCP connections of your tools, like QuickBooks, your CRM and your email platform, and it can gather and combine data from all of them. Your AI becomes the single place you run the business from.

8:44 Why The AI Sales Rep Stopped Working In Month Two

The story sales people argued about all week: one operator ran an AI sales rep for a client. In month one it booked a stack of calls. In month two it booked nothing, with the same tool, lists and messages, because prospects had learned to recognize the pattern. Volume was the whole promise, and volume is what gave it away. Chris’s lesson is freshness. People are happy to deal with AI that answers right away and solves their problem, as long as it does not feel like a repetitive robot.

12:08 Why The Hot New Model Suddenly Got Worse

OpenAI’s GPT-6 Astra launched to strong reviews, including a 3D architectural fly-through Chris calls professional grade. Then users reported it had suddenly gotten worse. His explanation is a pattern he has seen with every hyped launch: everyone rushes to a new model, the provider runs short of compute, and it quietly scales the model back rather than go down, because an outage costs paying customers more than a slightly weaker model. He says the same thing has happened at Anthropic after its launches.

14:52 Where Smarter Models Actually Pull Ahead

Chris does not buy the talk of a new era of general intelligence. What he does see in the top models is where they separate from cheaper ones. Coding and other procedural work are handled well by lower-end models. The difference shows up in creative work: distinctive websites, 3D renders, animation in tools like Blender. That is the only real separation he sees, and he did not expect it.

17:00 The First AI Music Tool With A Paper Trail

Suno launched version six: three models built with Warner Music Group, BMG and Believe, trained on licensed music, with every older model retired the same day. For anyone putting music behind client videos, it is the first major AI music tool with a paper trail. The other side: Suno’s own court filing admits it scraped YouTube to train its earlier models, and Universal, Sony and Round Hill are still suing. Chris expects creators to end up generating bespoke music for each project rather than dealing with licensing at all.

19:56 The $20 Agent That Books, Buys And Sends For You

Meta launched Muse, a personal agent that runs in its own cloud sandbox and completes tasks: it books, buys, drafts and sends. It has its own app and works inside WhatsApp, for adults in the US, with a free tier, a $20-a-month tier and a $100-a-month tier. For the full version it needs your inbox, your calendar and your card. Chris thinks this is how AI reaches everyday people, and that Meta is making the most practical moves of the big companies while getting the least attention.

22:13 Why Small Businesses Never Get An Operations Team

Most small businesses do not have the revenue for an operations team, so the owner does it all. Even when Chris’s own business could afford it, he held back. Hiring operations people is a low-odds bet, good ones leave for better jobs, and becoming a full-time manager of people changes who you are as an owner. Action-oriented owners would rather launch things. His picture of what AI changes: the owner as an orchestra conductor, pointing at each section and bringing it in.

26:43 What An AI Employee Actually Is

Following the episode slides, Chris builds on the integrator Olga introduced the week before: one AI you talk to, with a team of operators under it. Then he answers the question of what these employees are. They are a little data and some files. Each one has three layers. The seat is a small data file with the role, title, what it owns and how success is measured. The jobs describe the work in detail. Under those are the procedures. A single role, he says, turns out to be surprisingly little data.

31:26 Procedures Are Atoms. Workflows Are Molecules.

A procedure happens inside one role, and nobody else needs to know about it. A workflow is owned by the business: something comes in, one role does its procedure, passes the output to the next, and so on until the business reaches an outcome. What businesses actually do is almost always a workflow. Assign the procedures to the AI employees, and the org chart is simply all of them together.

33:43 A Real AI Sales Director, Mapped Out

Chris shows a sample org chart from Olga’s business. She sits at the top and talks to Midas, her AI director of sales. Midas owns procedures under prospecting (ideal customer fit, lead generation, trigger events, client news), meeting prep (triage, discovery prep, call debriefs), CRM updates, action items, follow-ups, the pipeline and a daily report. Anything that needs her sign-off or decision comes back to her.

35:17 Your AI Employee Reads Its Job Description Every Time

A person reads the job description once, on day one. An AI employee reads it every single time it does anything. Run that a thousand times and a person drifts while the AI reinforces the same role, limits and rules every run. In operations, consistency is what you want and inconsistency is what causes problems. Chris is blunt that human employees bring risk as well as value, and that AI will take a lot of that risk out of a business.

38:21 The Layer Nobody Wants To Write

The procedure layer is where every task is written out in full so an AI can follow it. It is exactly what real businesses lack. Employee handbooks get read in week one and never again. With AI, every task is spelled out, and the AI follows it like a champ. The result is a detailed job description for each operator, with guardrails for what it does and does not do.

40:28 AI Isn’t Killing Software. It’s Turning Your Business Into It.

The popular idea is that AI kills software because everyone will build their own. Chris calls that the pebble in front of the real story. Businesses themselves are becoming software: employees, org charts and every repeated process. As that happens, it becomes clear where people are needed: relationships, and the spark of ingenuity that moves a business forward. He sees that as freedom. Nobody gets value from spending the day on the same repetitive paperwork, whatever they are paid.

44:43 Why Renting AI Employees Is A Red Flag

Chris predicts a wave of companies selling AI employees as a monthly subscription, maybe $500 a month for one role and $300 for another. His objections: yet another subscription on top of the ones you already have, employees from different vendors that will not talk to each other, and the same connection work repeated for each one. Running them on one computer in your office does not solve it either once other people need access, and a shared Dropbox folder is risky.

47:16 The One Place Online You Already Own

The best place for your AI employees is one place online that you own, and that is your website. Chris calls the set of files a brain: the employees, their jobs and the workflows. The brain is inert until you point an AI at it. His image is Krang from the 1987 Teenage Mutant Ninja Turtles: a brain that can do nothing on its own until it is put in a body. The AI is the brain and your business files are the suit. By default the AI starts as your integrator, and you ask it what needs your attention. PageMotor already supports this, and Chris is working with its beta users on it.

55:36 Your Business Is Now A Data Structure

In software, you define the data structure before writing code, and the code follows its shape. Chris argues that setting up AI employees is the same step for a business. The data files are the structure, the markdown instructions are the code, and the result is the business. He says it makes a business that used to feel enormous seem small once it is written down as data.

58:18 Four Rounds Took 80% Of $12.4 Billion

AI raised $12.42B across 71 companies from September 3 to 9, 51.7% of all venture dollars. Four rounds were 80% of it: Mistral, $3.49B, Paris, AI models and infrastructure for governments and companies that want a European alternative. Crusoe, $3B, Denver, AI data centers and the power that feeds them (reported). Cognition, $2B, San Francisco, maker of Devin, an AI that writes and ships software on its own. Fluidstack, $1.5B, New York, GPU clusters for companies training models (reported). Also in the top five: Harvey, $550M, San Francisco, AI for legal work (reported). Strip out the four biggest and the rest of the AI market raised about $2.43B, roughly half the week before. Chris is openly skeptical that AI can yet design software the way an experienced engineer does, and says the proof would be visible if it could. The full Week 41 report is here.

1:09:04 The Control Panel For Your Whole Business

Chris closes with the villain from the 1980s cartoon Danger Mouse, who runs his empire from one chair and one control panel. That is how he pictures the owner’s relationship with AI: one interface, connected to your CRM, payments, accounting, legal, sales and marketing, answering any question and surfacing what needs you. Instead of logging into a dozen tools, you talk to one AI, on your phone or at your desk. He expects a big fourth quarter for agents, and AI running businesses to be where 2027 heads.

Resources And Sources

• Your First AI Employee, the slides.
• NBC News Decision Desk poll, September topline. The 52%, 70% and 18%.
• OpenAI on the ChatGPT data agent.
• Jakub Pachocki, An Alien Mind. OpenAI’s chief scientist on why no lab has solved alignment well enough to keep scaling.
• Suno on version six.
• Meta on Muse.
• AI Funding Report, Week 41.
• Weekly AI Funding Tracker. 41 weeks.
• Episode 57. The AI employee, running live.
• PageMotor. The AI-native CMS.
• The email list. PageMotor and Practical AI updates.
• Midas Org Chart. What an AI-run sales department looks like.
• The Job Description You Write For A Machine.
• One Job, Written Down.