How to Build Your First AI Agent: A Single-Session Guide for Mid-Career Pros
One sitting, five moves, one working agent pointed at a real task you already have this week.
You build your first AI agent by giving it a name and a job, writing what it should do in plain language, switching on the specific abilities that job needs, connecting it to the tools it has to touch, and running it once on a real task before you trust it with a second. On a no-code platform, that sequence takes about an hour. There is no code to write and no engineering ticket to file.
The part most people get wrong is scope. They try to design the agent that runs their whole week, and they never finish. Build the one that handles a five-minute task you repeat, get it running end to end, and expand its job only after you have watched it work.
An AI agent is a chatbot with tools and instructions. It reads context, follows a written brief, and takes action inside the software you already use, instead of only answering questions in a chat window.
What do you need before you sit down?
Three things, and none of them are technical.
An account on a no-code agent platform, opened before you start. Jacqueline teaches these builds on Gumloop, which runs on a 14-day trial and then a paid plan. Set it up the night before. That hour is for building, not sign-up screens.
The login for whatever your agent has to reach. For most first agents that is a calendar or an inbox. Have the account you plan to connect already signed in.
One real task you have coming up this week. A meeting to prepare for. An inbox to sort. A weekly report you assemble by hand. A first agent built against a real task sticks. One built against a hypothetical gets abandoned by Thursday.
What are the five moves that build any agent?
Every agent is the same five moves in the same order. This is the sequence used in the live Agent Lab sessions, and it is the sequence you reuse for every agent after this one.
- 01
Name it.
Give the agent a name and a one-sentence job. "Meeting Prep" is a name. "AI helper" is not.
- 02
Tell it the job.
Write the brief in plain words, the way you would brief someone on their first day. What it does, what it should never do, what good output looks like.
- 03
Switch on its abilities.
Turn on only the skills the job requires. Web search. Drafting email. Reading a calendar. Nothing beyond the job description.
- 04
Connect your tools.
Give it scoped access to the calendar, inbox, or folder it needs. Scoped access to the thing it works on, not the keys to your whole account.
- 05
Run it.
Watch the first run from start to finish. Then point it at the real task you picked before you sat down.
Move two is where most first builds fail, and it fails quietly. A vague brief produces vague output, and the output looks plausible enough that people blame the model. Write the brief the way you would write it for a new hire who cannot ask you a follow-up question.
What should your first agent actually do?
Pick something narrow and recurring over something ambitious and rare.
A Meeting Prep agent is the standard first build in her sessions: it reads the calendar, pulls the attendee list and the last thread with those people, and hands you a brief before the call. Almost everyone has a meeting this week to test it on, which is the whole point.
Participants in the live sessions have built variations on the same idea. One host of several companies built a separate meeting-prep agent for each one, each reading its own calendar and inbox, because the businesses were too different to share a brief. Another built a Scout agent that searched executive-search firms inside a set radius on a schedule, and only after she had tested it by hand.
Notice the shape of both. Narrow job, real data, tested before it was trusted.
What rule keeps a first build safe?
Draft, never send.
That instruction goes in the brief before the agent touches anything that leaves the building. The agent writes the email and puts it in your drafts folder. You read it, you fix it, you send it yourself. You do that a dozen times before you consider letting it act without a review step, and plenty of people who run agents daily never remove that checkpoint at all.
The same discipline applies to when it runs. Keep the trigger on manual until you trust the output. Move it to a schedule after you have seen a week of runs you would have been comfortable sending.
BNEDai's AI Readiness Index scores teams on eight of these operating habits, including scoped access, graduated trust, stop-and-ask behavior, and reversibility. Every one of them is a decision you make while you build the agent, before anything runs.
How long does the first build take, and what about the second?
In a guided session it fits inside a single hour, with a working agent at the end instead of a slide deck about agents. Building alone the first time, block ninety minutes. The five moves take the same time either way; you'll spend the extra half hour checking your own work, which is the correct instinct.
The second agent is much faster. You already know what a brief looks like, what scoped access means, and where the platform hides the trigger settings. Most people build their second one in a fraction of the time and stop thinking of it as a project.
When does one agent become several?
When the job has clearly separate steps that need different instructions.
The pattern she teaches for that is a manager agent with specialists under it. In her job search agents build, one Job Search Manager holds your profile and hands work to three sub-agents: Scout finds fresh roles on company career pages, Matcher scores each one against your profile and names two reasons a hiring manager might pass, and Ghostwriter drafts the outreach from real samples of your own writing.
That architecture transfers. Any workflow where the finding, the judging, and the writing are three different jobs can be split the same way. The one thing that has to come first is a working single agent. Build one that runs before you build four that talk to each other.
What to do this week
Open a platform account tonight. Block sixty minutes. Pick the meeting you're least prepared for, and build the agent that prepares you for it. Keep it on draft-only and manual until you have watched it work.
If you would rather do that first hour with someone in the room, the Agent Lab sessions are free and run live, and you build your own agent alongside everyone else rather than watching a demo of one.
Related reading: What Is an AI Agent? covers the definition in under 600 words. What Is an Agent Build Session? explains the format before you walk in. AI Agent Training for Corporate Teams covers the same build as a capability audit for a whole team. AI Workshops for Business is the cohort version, a group building on a schedule rather than one person building once. Team AI Workshops covers running the session yourself for your department. Managing AI Anxiety is for the part of this that is not technical at all. AI Upskilling for Professionals is the strategic shift from building one agent to managing a personal portfolio of them.
Frequently asked questions
Do I need to know how to code to build my first AI agent?
No. All five moves are done in plain language on a no-code platform. You name the agent, describe its job the way you would brief a new hire, switch on the abilities that job needs, connect the tools it has to touch, and run it. Nothing in the sequence requires writing a function or filing an engineering ticket.
How long does building a first AI agent actually take?
In a guided live session it fits inside a single hour, start to finish, with a working agent at the end. Building alone for the first time, block about ninety minutes. The extra half hour goes to checking your own work. Your second agent takes a fraction of that, because the five moves are already familiar.
What should my first AI agent actually do?
Something narrow and recurring rather than ambitious and rare. A Meeting Prep agent that reads your calendar, pulls the attendee list and the last thread with those people, and hands you a brief before the call is the standard first build, because almost everyone has a real meeting this week to test it against.
Is it safe to let an agent send email on my behalf?
Not on day one. Put "draft, never send" in the brief so the agent writes into your drafts folder and you send it yourself. Keep the trigger on manual until you have watched a week of runs you would have been comfortable sending. Graduated trust and reversibility are two of the eight habits the AI Readiness Index scores.
What does it cost to build one?
The live BNEDai build sessions are free. The platform is the cost: Gumloop runs on a 14-day trial and then a paid plan. Budget for the tool, not for the session, and decide whether to keep paying after you have seen your first agent run for a week.
Build something that actually runs your workflow.
A focused, free 60-minute live session with Jacqueline. You build alongside her, on your own real task, and leave with an agent that is already running.