Onboard an AI agent the way you'd onboard a new hire.
Management skills are AI agent skills. The six things you give a person on day one are the six things an AI agent needs before its first run. Here is the sheet Jacqueline V. Twillie fills out before she builds one.
By Jacqueline V. Twillie, MBA. Founder of BNEDai. From her conversation on Office Hours: The Black Girl Podcast, September 28, 2026. Updated October 19, 2026.
The test
If you hired a brilliant new team member today, handed them zero context, locked them out of your systems, and yelled a one-sentence command across the room, they would fail. That is how most people set up an AI agent.
An AI agent is a workflow connected to your systems that runs on instructions. If you want it to deliver, onboard it the same way you would onboard a competent new hire. Management skills are AI agent skills.
The onboarding sheet
Six parts. Fill every one before you open the build tool. Each part is a question a good manager answers for a person on their first day.
Job title and goal
If this AI agent succeeds this week, what one measurable problem did it solve? One outcome, in one sentence. An AI agent with three jobs does none of them the same way twice.
Role and context
Who is this AI agent acting as, and what does it need to know about how you operate? Give it a role (meeting prep, revenue research, UX review) and the background a person in that role would get on day one: who you serve, what good looks like, what you never do.
System access
Which tools does it have permission to read or write in? Name them: Slack, Airtable, the calendar, a dedicated email account for AI tools. Give it the fewest it needs and nothing it does not.
Step-by-step instructions, and the why
What exact steps should it follow, and why does each step matter? Tell it what to do, what not to do, and the reason. A person who knows why a step exists handles the case you did not write down. So does an AI agent.
Guardrails
What is this AI agent never allowed to do without a human yes? Send, delete, pay, publish, promise. Write the limits before the first run. Nothing leaves without your review.
Reporting and review
How and where does it report back, and how will you judge whether it saved you time? Pick a format and a day. Jacqueline reviews her AI agent team on Fridays and asks which ones moved the week's goals.
Write your first AI agent job description in five minutes
Pick one repetitive task you dreaded this week. Moving information into forms. Prepping the weekly status notes. Then fill in these five blanks in your notes app.
- I am hiring an AI agent to serve as my [role].
- Its number one goal is to [specific outcome].
- It will only have access to [one or two tools or documents].
- Its boundary is: never [action] without my review.
- Every [day and time], it will send me [deliverable] so I can check its work.
That is the brief. Bring it to a free Agent Lab session and build it in about an hour, or follow the answers on building your first AI agent.
What is the new hire analogy for AI agents?
The new hire analogy is Jacqueline V. Twillie's test for setting up an AI agent: give it what you would give a competent person on their first day. If you hired someone today, handed them no context, locked them out of your systems, and shouted a one-sentence instruction across the room, they would fail. Most people set up an AI agent exactly that way, then blame the AI agent.
An AI agent is a workflow connected to your systems that runs on written instructions. The skill it takes to set one up well is management, not programming. Define the job, give the context and the access, set the guardrails, and review the work.
What goes on an AI agent onboarding sheet?
Six things: the job and its one measurable goal, the role and the context behind it, the systems it may touch, the step-by-step instructions with the reason for each step, the guardrails it never crosses without a human yes, and how and when it reports back for review. Fill all six before you build. Each one maps to something a manager already does for a new person.
Why does an AI agent need to know why a step matters?
Because the why is what handles the case you did not write down. A person who knows the reason a step exists makes a sensible call when the input looks different. An AI agent given only the step repeats it, wrong, in every case you did not anticipate. Jacqueline tells every AI agent what to do, what not to do, and why.
What is a guardrail for an AI agent?
A guardrail is a written limit on what the AI agent may do without a person's approval. The ones Jacqueline sets first: it never sends, deletes, pays, publishes, or promises on its own. In her live builds no connectors are turned on, so the AI agent drafts and the person sends. Delegating the execution of a task never removes your responsibility for the decision.
How often should you review an AI agent's work?
Weekly, on a fixed day, the way you would check in with a new team member. Jacqueline sets the week's goals on Monday and reviews her AI agents on Friday: which ones she used, what they produced, which sat idle, and which moved a goal forward. An AI agent that never gets reviewed drifts, and you find out when something ships wrong.
How do you write your first AI agent job description?
Pick one repetitive task you dreaded this week, such as moving information into forms or prepping weekly status notes, and fill five blanks: the role, the one goal, the one or two tools it may touch, the thing it never does without your review, and the day and format it reports back. That takes five minutes and gives you the brief for a first build.
Build one live.
Bring your onboarding sheet to the next free Agent Lab session, or bring Jacqueline to your team.