How to Write a Job Description for an AI Agent

The fastest way to tell whether an AI agent project will work is to ask for its job description. If the answer is a prompt, the project is an experiment. If the answer is a role with a trigger, an outcome, access rules, and a manager, the project is a hire.

Most teams skip this step because the agent is software, and software gets requirements, not job descriptions. But an agent that acts inside your business is closer to a new hire than to a new app. It needs to know what it is for, what good looks like, and who to go to when it is stuck. Gartner's 2025 prediction that more than 40 percent of agentic AI projects will be canceled by 2027, partly for unclear business value, is what skipping that step looks like.

An AI employee is an AI agent with a job, and the job description is where the job gets written down. Here are its eight parts, with a worked example.

Role and trigger: what it is for, and when it starts

Role is one sentence. "Welcome every new customer and get them to first value." "Answer every inbound call and book the appointment." The role sets the boundary: what the agent owns and, by omission, what it does not. An agent with two roles has none. Vida's six agent types are role definitions: AI Receptionist, Lead Conversion, Lead Recovery, Customer Activation, Customer Service, and Retention and Renewals. Each owns one stage of the customer relationship and hands off to the next.

Trigger is the event that puts the agent on the clock. People start work when they arrive. Agents start when something happens: a call rings, a form is submitted, a customer signs, a renewal date is 30 days out. It should be specific enough that two people would agree whether it fired.

Outcome: what done looks like

This is the line most agent job descriptions are missing, and the one that matters. The outcome is the business result the agent is responsible for, with a number attached. Not "handle onboarding" but "onboarding completed within seven days for 80 percent of new accounts."

The research says why. McKinsey's 2025 State of AI survey found the organizations capturing significant value from AI were nearly three times as likely to have redesigned workflows around it, and pointed to KPI tracking as part of the difference. A job description with a number in it is the smallest possible workflow redesign.

Skills and access: what it can do, and where

Skills are the specific things the agent is allowed and able to do: send a welcome message, collect a missing document, schedule training, update a milestone field. List them. That list is the agent's starting scope, and it grows only when the manager adds to it. Agents do not pick up skills on their own, and the job description should not imply they do.

Access names the software the agent logs into and what it can read and write there. CRM: read contact and account, write milestone status. Calendar: read availability, write bookings. Documents: write uploads. Anything not listed is off limits. This is where risk control starts. The NIST AI Risk Management Framework puts Govern and Map ahead of Measure and Manage for the same reason: define the scope before you measure the behavior.

Rules and handoffs: the boundaries

Rules are the policy layer: the approved answers, the questions it must not answer, the hours it may send messages, the consent rules it follows, the price it never quotes. Write them as plainly as you would for a new hire.

Handoffs cover what the role does not. Define them, and what each handoff carries: conversation history, systems touched, steps already taken. A handoff is not a failure. It is part of the job, and the person receiving it should never have to ask the customer to start over.

Manager: who reviews the work

A named person on your team. They read the scorecard, review conversations, approve new skills, and decide when the agent takes on more. Without a manager, the other seven parts drift. With one, the agent is an employee. Capgemini's 2025 survey found 38 percent of organizations expect AI agents to be members of human teams by 2028, and a team member reports to someone.

The worked example: AI Customer Activation Agent

Role: get every new customer from signup to first value. Trigger: a customer signs, pays, or creates an account. Outcome: onboarding completed within seven days for 80 percent of new accounts, measured monthly. Skills: welcome within minutes, explain the next step, collect missing items via secure links, schedule training, confirm milestones, detect stalls. Access: CRM contact and milestone fields, calendar, document uploads. Rules: approved resources only, no contract or billing changes, outreach inside business hours. Handoffs: any blocker the agent cannot resolve in two attempts goes to the onboarding lead with the full thread. Manager: the head of customer success.

Eight lines. That is a job, and the agent that fills it is a digital employee you can hold accountable.

Hiring on Vida

Vida is where businesses hire, manage, and measure AI agents as digital employees. A Vida agent is built from the job description above: a role from the six agent types, a trigger, outcome KPIs, plain-language skills, scoped access through its own browser, rules, handoff paths, and a manager on your team. The AI workforce management layer keeps every part visible and versioned as the job grows.

Bring the role you would hire for next. We will help write the job description and prove the outcome in a paid pilot on live traffic. Design Your AI Agent.