Hiring Your First AI Employee: Notes From Greg Brainerd's Session in Spooner
An AI employee for small business, in plain English: pick one job, limit what it can touch, set approval gates, and review it in 15 minutes a week.
In this article
- What is an AI employee for a small business?
- What Greg Brainerd's session in Spooner covered
- Pick one job for your first AI employee
- Write the job description in plain English
- Permissions and approval gates for an AI agent
- "No developers" does not mean no code
- What to delegate first, and what never to delegate
- The 15-minute weekly review that keeps it honest
- Frequently asked questions
- Talk to us before you hire your first AI employee
An AI employee for small business is an AI agent with one defined job, access to only the tools that job needs, and a named person who checks its work. It is not a new hire, and it does not replace one. It is software that can read your data, draft the next step and, within limits you set, take that step. The practical first hire is narrow: a weekly report, a follow-up list or a scheduling chore, with read-only access to start and human approval before anything leaves the building.
Greg Brainerd's session at the AI Business Retreat in Spooner, Wisconsin, was built around that shape: one job, clear limits on what the agent can read, and a 15-minute weekly review. Below is what the session covered, plus the permissions, approval gates and review habits that make a first AI employee safe to run.
What is an AI employee for a small business?#
If you have used ChatGPT or Copilot, you have used an AI assistant. You ask a question, it answers, and you decide what to do next. An agent takes a goal instead, works out the steps, uses connected tools to carry them out, and reports back. Our plain-English explainer on agentic AI covers that shift from AI that answers to AI that acts, and our guide to the difference between an AI assistant and an AI agent shows where each fits in your business operations.
"Employee" is a useful word for it because it forces the right habits. You would not hire a person without a job description, a login with the right access, and a manager. An AI employee needs the same three things.
What Greg Brainerd's session in Spooner covered#
The AI Business Retreat 2026 was held September 15 at The Island in Spooner, with three tracks: Leadership, Sales and Marketing, and Builder. Hire Your First AI Employee (No Developers, No IT Department) was scheduled for 10:00 to 10:30 AM on the Leadership Track. It was written for owners and executives buried in reporting, follow-up or scheduling who have no developers on staff.
The presenter was Greg Brainerd, a Certified AI Business Maturity Model (AIBMM) Coach and the founder and CEO of Braintek in Houston. The outline worked through six points:
- Pick the job: reporting, follow-up or scheduling.
- Describe it in plain English and watch AI build it, live.
- What "AI builds it" really means: there is code, and someone has to own it.
- What to delegate first, and what never to delegate.
- The 15-minute weekly review that keeps it honest.
- What it can read and what it can't: guardrails set in the build, not after.
The sections below turn each point into practical steps. For the rest of the day, see our photo recap of the AI Business Retreat 2026.
Pick one job for your first AI employee#
The session named three starter jobs. Each is frequent, repetitive and easy for a person to check.
| Starter job | What it reads | What it produces | Where a person approves |
|---|---|---|---|
| Weekly reporting | Sales, jobs, invoices or tickets | A one-page summary of the numbers and what changed | Before it goes beyond the leadership team |
| Follow-up | Your CRM, open quotes and recent inquiries | A daily call list with drafted messages | Before any message is sent |
| Scheduling | Calendars, a job board or a service queue | Proposed times, reminders and conflicts | Before anything is booked with a customer |
Pick the job that costs you the most hours and has the lowest cost of a mistake. Internal reporting is often the safest start, because a wrong number gets caught before anyone outside the company sees it.
Write the job description in plain English#
The "no developers" promise rests on this step. You describe the job in plain English and AI does the building, so the quality of the description sets the quality of the employee. Brief it like a new hire:
- The job: one sentence, such as "Every Monday at 7 AM, summarize last week's closed jobs and open quotes."
- The inputs: which systems, folders or reports it may read.
- The output: exactly what it hands back, and to whom.
- What good looks like: two or three examples of a correct result.
- What it must never do: send, delete, pay, promise or change anything without approval.
- Its manager: the person who reviews its work every week.
If you cannot write item 4, the process is not ready for an agent. Fix the process first.
Permissions and approval gates for an AI agent#
This is the step that is easiest to skip, and it is where the risk lives. The session's point was that guardrails get set in the build, not after. Think of access in three levels.
| Level | What the agent can do | Example | Approval |
|---|---|---|---|
| Read | Look at data | Pull last week's invoices for a report | Not needed for internal use |
| Draft | Prepare an action without taking it | Write a follow-up email and leave it in drafts | A person reviews and sends |
| Act | Take the action itself | Send the email, book the slot, update the record | Low-stakes steps only, after a track record |
Start every AI employee at Read and Draft. Move a single task to Act only after it has earned it, with weeks of drafts that went out unchanged.
Security people have a name for getting this wrong. The OWASP Gen AI Security Project's 2025 Top 10 for LLM applications lists Excessive Agency (opens in a new tab), caused by excessive functionality, excessive permissions and excessive autonomy. Its guidance: limit what an agent can call and reach to the "minimum necessary," and require a human to approve high-impact actions before they are taken.
OWASP's own example fits a small office. An assistant gets mailbox access to summarize incoming email, but its tool can also send mail. A maliciously crafted incoming email tricks it into scanning the inbox for sensitive information and forwarding it to the attacker. That trick is called indirect prompt injection (opens in a new tab): instructions hidden in outside content the AI reads. The fixes OWASP lists are the same ideas as the table above: a tool that only reads mail, read-only mailbox access, and a person who reviews and hits send on every draft.
In practice, human approval belongs in front of:
- Anything that leaves the building: emails, texts, quotes and posts to customers, vendors or the public.
- Money: payments, refunds, credits and pricing.
- Changes you cannot easily undo: deleting files, editing customer records, cancelling appointments.
- Commitments: contract terms, delivery dates and promises of any kind.
Give the agent its own login with its own limited access, not yours. Then you can see what it touched and switch it off without changing your own password.
Our own internal AI employees follow the same rule: actions that write or change anything stay behind human approval. For the details, see how we built our AI staffing department.
"No developers" does not mean no code#
The session made a point that is easy to miss: when AI builds your agent, there is still code behind it, and someone has to own it. Before your first AI employee goes into daily use, write down:
- Whose account it runs under, and where it runs.
- Who holds the passwords and connection keys it uses.
- What happens when a connected app changes its screens or its data.
- Who fixes it when it breaks on a Monday morning.
Building the first version is the quick part. Error handling, drift as your process changes, credentials and updates are the ongoing work, and they need a name next to them.
What to delegate first, and what never to delegate#
Delegate first: work that is frequent, repetitive, easy to check and cheap to get wrong. Summaries, first drafts, reminders, sorting and list-building all fit.
Keep with people: hiring, pay and discipline; pricing and contract commitments; moving money; conversations with an upset customer; and any call you would be uncomfortable explaining as "the software decided." An agent can prepare the file. A person makes the call.
The 15-minute weekly review that keeps it honest#
Put it on the calendar of the agent's manager, at the same time every week:
- Spot-check five outputs. Are the numbers right? Would you have sent those drafts?
- Read the activity log. Did it touch anything outside its job?
- Check its access. Remove any permission it did not use.
- Look for drift. Has your process, price list or software changed since you wrote the job description?
- Decide. Keep it, adjust the instructions, move one task up to Act, or retire it.
An AI employee nobody reviews is an employee nobody manages.
Frequently asked questions#
Can a small business build an AI agent without developers?#
Yes, for a first agent with a narrow job. Many current AI tools can build a first version from a plain-English description of the job, its inputs and its outputs. You still need someone who owns the result: where it runs, the accounts it uses, and who fixes it when a connected system changes.
Will an AI employee replace my staff?#
A well-scoped AI employee takes over a chore, not a person. It does the gathering, sorting and drafting so your people spend their time on customers and decisions. A good first job is one nobody on the team wants anyway, such as assembling the Monday report.
What should an AI agent never do without human approval?#
It should not send anything to a customer, vendor or the public, move money or change prices, delete or overwrite records, or commit your business to terms or dates. OWASP's security guidance recommends human approval for high-impact actions and limiting an agent's access to the minimum it needs.
What is the difference between an AI assistant and an AI agent?#
An assistant answers when you ask and leaves the next step to you. An agent takes a goal, works through the steps with connected tools and reports back. That ability to act is why an agent needs permissions, approval gates and a weekly review that an assistant does not.
Talk to us before you hire your first AI employee#
Choosing the job, setting the permissions and deciding who owns the code shape everything that follows. Our AI adoption consulting starts by finding where AI would remove real hours in your business, ranks the candidates by effort against benefit, then helps you run the pilot and train the people who will use it every day. If you would rather learn hands-on first, start with our AI training for business, including the free monthly AI workshop.
