AI Employee: Write the Job, Then Hire It

An AI employee is an AI agent with a job: a role written in plain words, the knowledge and tools it may use, rules it follows, and a person it reports to. When you are also the front desk, its first job is the questions that arrive while you work. This guide takes the idea literally: write the job description, give it access, name its manager, compare the cost with a part-time hire.

What is an AI employee?

An AI employee (also called an AI teammate or digital employee) is an AI agent with a job: a role written in plain words, the knowledge and tools it may use, rules it follows, and a person it reports to. The term is used loosely: vendors apply it to phone bots, to teams of back-office agents and even as a product name. Unlike a scripted bot, it acts: it works out a price, checks a calendar, takes a request, and knows when to stop.

Ultimo Bots is the AI agent platform where you describe the job in plain words and Mo, the builder in the AI Studio, builds the agent that does it: skills that quote from your price rules, book in your calendar app or call your own systems through their API. One agent works on your website, your chat link, Instagram, Facebook Messenger and WhatsApp, with one set of knowledge and rules.

Will an AI employee take anyone's job?

Not the parts of a job that need a person. An AI employee takes the repeat questions and the simple, rule-based actions; the yes on an exception, the advice, the judgment call and the relationship stay with people. People ask which jobs will survive AI because they fear the whole job goes. At a front desk the whole job is rarely the problem: the evening flood of "how much for a three-bedroom?" is.

People rightly hate being stuck with a bot that will not let them through. A good AI employee knows where its job ends because you wrote it down, and it passes the conversation on: to you in the live chat while you are online, or as a request with the full conversation while you are not.

How do AI employees work?

An AI employee works from four things you give it: what it knows, what it can do, how it behaves, and who it hands over to. Lena's example below shows where each one lives in the AI Studio.

On every message the agent reads the conversation, picks the skill that fits, uses its tools and answers in the visitor's language. For how this fits into customer service as a whole, see our guide to AI customer service software.

Write the job description before you hire

Before you hire an AI employee, do what you would do for a person: write the job description. It forces the decisions that make or break the first week: which questions it owns, which actions it may take, what it must never promise, and who picks up when it is out of its depth. The best AI employees get a narrow job, written down. Pick the duties from last week's messages: the questions you answered more than once whose answer already sits in your price list, website or calendar. Anything that needed your judgment goes under "Never" or "Reports to" instead. Start there, and add a duty once the agent handles the first ones the way you would.

Still choosing the first job? Our guide to AI agents for business compares the options, and the AI receptionist guide covers the receptionist role.

Take Brightside Cleaning Co., a fictional home-cleaning company with three crews. Its owner, Lena Ortiz, is also its front desk. Quote requests arrive in the evening while she is on a job site or asleep, and by morning some of those people have booked whoever answered first. This is the job description she wrote:

Role: Quotes and bookings coordinator, Brightside Cleaning Co.
Where: our website, Instagram and WhatsApp, at any hour.

Duties:
- Answer questions about our services, service area and policies.
- Quote standard and deep cleanings from our price rules:
  $110 for up to 2 bedrooms and 1 bathroom, $25 per extra bedroom,
  $20 per extra bathroom, deep clean costs half again.
- Show open times in Cal.com and book the one the customer picks,
  after they confirm.

Never:
- Quote move-out or after-renovation cleanings. Offer to pass them to Lena.
- Promise a specific cleaner, or discuss refunds or damage claims.
- Pass as a person. If no reply so far has said so, say you are
  Brightside's AI assistant.

Reports to: Lena. When she is online, hand over if someone asks
for a person. When she is not, take the request for her.

Every line has a home in the agent. The role and the facts become its Business Context, the questions it answers come from its knowledge, the actions become skills, the "never" lines become behaviors, and "reports to" becomes the handover.

Hand the job description to Mo

In the AI Studio, the job description is the setup: Lena pastes it into the chat with Mo, and Mo builds a skill for each action it names. Mo asks when something is unclear.

For quotes, Mo builds a skill with a small calculation from Lena's price rules. Mo tests the calculation before it reports back, so a three-bedroom, two-bathroom house comes out at $155 every time, the same answer Lena would give.

For bookings, Lena connects Cal.com, where her crews' cleaning slots live: Mo shows a Connect button, and she signs in on the Apps page, with no API key. Mo then adds the ready-made skill "Book meetings in chat", which shows free times of a Cal.com event type, asks for the customer's name and email, and books the slot after the customer confirms. Ready-made skills also exist for Wix, Calendly, Stripe, HubSpot and Mailchimp.

The "never" lines become behaviors. Mo words each as a rule of up to 250 characters and saves it, and Lena can read and edit every rule in the Behaviors tab of the "Your agent" panel.

If Lena kept her schedule in her own booking system with a public API and an API key, Mo would build the same skill on that API: it reads the vendor's API documentation, asks for the key in a secure field, stores it encrypted and tests every tool it builds. The same works with MCP servers.

Here is the result on a Tuesday at 10:40 PM, in the chat on Brightside's website. A visitor writes: "How much is a regular cleaning for a 3-bedroom, 2-bath house?" The agent answers: "I'm Brightside's AI assistant. A standard cleaning for 3 bedrooms and 2 bathrooms is $155. Would you like to see the next open times?" The visitor asks for Thursday. "Thursday has 9:00 AM and 1:30 PM free. Which one works for you? Then I just need your full name and email." Once the visitor has picked a time and said yes, the cleaning is booked in Cal.com, and Cal.com emails the confirmation.

Day one: the keys an AI employee gets

Give an AI employee only what the job needs, the way a new hire gets a key to the front door but not to the safe. Lena's list:

• Her content: website pages, a price sheet as a PDF and a few Q&A pairs; the knowledge base also takes Google Drive, OneDrive and Notion.
• One connected app: Cal.com, signed in with its own login. A connection only makes an app available; the agent uses it only in a skill that includes it.
• Customers' own bookings: before the agent shows or moves a customer's booking, the customer confirms their email with a 6-digit code. Lena added the ready-made skill "Move or cancel bookings" for that.
• Money: card details never enter the chat. If Lena took deposits, a Stripe skill would hand over a secure payment link.

Who is the human behind your AI employee?

Its manager is you, or whoever on your team is online, and the handover is how you manage it in the moment. Teammates are free in Ultimo Bots, so the whole crew can be in the live chat.

While Lena or a teammate is switched online in the portal, a "Talk to a human" button shows in the website chat. When a visitor clicks it or asks for a person, the teammate joins the same window with the whole conversation and the agent stops answering until the teammate leaves.

At 3 AM nobody is online, so the button stays hidden and the agent keeps working. A visitor asking about a move-out cleaning hears that Lena quotes those herself; the agent offers to pass the request on, asks for name and email only after a yes, sums the request up and confirms that the team has it. The request lands in Leads with the full conversation, and the new-lead email goes to Lena, or to every address she lists under Email recipients in Notifications, her own included.

In the morning Lena can also reply to a website chat that has already ended: the visitor sees her reply when they come back to the website. On Instagram, Facebook Messenger and WhatsApp she pauses the AI for one conversation and replies from the portal Inbox. Our guide to human handoff goes deeper.

How much does an AI employee cost?

An AI employee on Ultimo Bots starts at $29 a month on Smart; a part-time hire for 20 hours a week costs about $1,866 a month in wages alone.

The wage comes from the U.S. Bureau of Labor Statistics: "The median hourly wage for customer service representatives was $21.53 in May 2025" (BLS Occupational Outlook Handbook, page updated October 29, 2026, read November 13, 2026). That is before payroll taxes, and only in the hours you schedule.

MonTueWedThuFriSatSunPart-time hire20 hAI employee168 h

Part-time hireAI employee

The numbers do not say the agent replaces a person; they say what an hour of cover costs. The AI employee covers all 168 hours of the week, but only the work you described. A person covers the hours you pay for and everything that needs judgment.

AI credits are the unit you pay in: a standard answer uses 1 AI credit, so Smart covers up to 500 standard answers a month. As of November 13, 2026, Smart costs $29 a month or $19 billed yearly with 500 AI credits, Boost $59 or $39 with 2,000, and Ultimo $249 or $199 with 10,000. There is no setup fee: Mo's work draws on the same balance, teammates cost nothing extra, and agents installed from Wix are billed through the Wix App Market. More in what an AI agent costs and the AI agent cost calculator.

Its first week: test it like a customer

Treat the first week like a trial period: test the agent the way a customer would, then read what real visitors asked. Before you switch it on, note how many quote or booking requests reached you after hours last week. At the end of the 7-day trial, count the new leads and bookings that came in while you were away, and compare.

1. 1

   Ask the questions you hear most

   In the Test tab of the AI Studio, ask your five most common questions, one your content does not answer, and one that should go to you. Test chats use AI credits, and a test booking is real, so cancel it in Cal.com.

2. 2

   Let Mo play the difficult customer

   Ask Mo: "Test my agent like a grumpy customer who wants a refund." Mo runs the conversation and reports how the agent handled it.

3. 3

   Read the real conversations

   After a few days, ask Mo: "Find questions my agent couldn't answer and close the gaps." Mo reads the real conversations and proposes what to add.

4. 4

   Update the job description

   Every answer you would phrase differently becomes a Q&A pair or a sharper rule. Tell Mo, as you would brief a new hire.

Your next 10 minutes

Write the job description before you open anything: one role, three duties from last week's messages, two things it must never do, and who it reports to. Then start the trial, paste it to Mo, and ask your new agent the question that reached you last night. If the answer is the one you would have given, your AI employee has started work.

Questions and answers