The 6-Box AI Agent Onboarding Canvas
One page, six boxes, written before you open the build platform.
The 6-box AI agent onboarding canvas is a one-page worksheet you fill out before you build: the job title and north-star goal, the persona and operating context, the system keys it gets, the standard operating procedure with the why behind each step, the guardrails, and the Friday report card. Each box is a question. Answer all six and the brief you paste into the build platform is already written.
This is the worksheet behind the new hire approach to onboarding an AI agent. The approach says what an AI agent needs. The canvas makes you write it down before you touch a tool, which is where most first builds go wrong. First builds tend to go wrong the same way: the platform is opened first, the clicking starts, and the result is with an AI agent that has access to everything and a job that fits in one vague sentence.
What goes in each box?
- 01
Job title and north-star goal.
If this AI agent succeeds this week, what single measurable problem did it solve? Write the title the way you would on a job posting ("Meeting Prep Assistant") and the goal as a sentence you can check on Friday ("Every external meeting on my calendar has a one-page brief in my drafts by 7 AM").
- 02
Persona and operating context.
Who is this AI agent acting as, and what does it need to know about your business to do the job the way you would? The persona might be a UX researcher, a revenue strategist, or an executive assistant. The context is the standard documents, the clients that matter, your preferences, the tone you write in. Everything a new hire would absorb in their first weeks goes here in writing.
- 03
System keys.
Which specific tools can it read, and which can it write to? List each one by name: Google Calendar (read), Gmail (read, write to drafts only), Airtable (read), Slack (none). A dedicated email address for your AI tool accounts belongs in this box too, so the AI agent is never connected to the inbox that holds your banking and medical mail.
- 04
Standard operating procedure, with the why.
The exact steps, in order, and why each one matters. "Read the attendee list. Pull the last email thread with each person, because the brief should pick up where the last conversation stopped. Draft one page. Save to drafts." The why is what lets the AI agent handle the case you did not anticipate.
- 05
Guardrails.
What is this AI agent strictly forbidden from doing without a person? Start with draft, never send. Add never delete, never reply to a thread it did not start, never touch anything outside the systems in box 3. Write them as rules, one per line.
- 06
Friday report card.
How and where does it report back, and how will you judge whether it saved you time? Name the deliverable, the place it lands, and the day you will read it. The first few Fridays, read everything. After that, read the exceptions.
The 6-box AI agent onboarding canvas is a pre-build worksheet. Six questions, one page: job and goal, persona and context, system keys, the procedure and the why, guardrails, and the Friday report. The completed canvas is the brief you paste into any no-code platform.
How do I use the canvas in five minutes?
Pick one repetitive task you dreaded this week. Transferring information into forms, prepping weekly status notes, building the same brief before every call. Then fill in five blanks in your notes app:
- 01
I am hiring an AI agent to serve as my [role].
Example: my meeting prep assistant.
- 02
Its number one goal is to [specific outcome].
Example: a one-page brief for every external meeting.
- 03
It will only have access to [one or two specific tools or documents].
Example: my calendar and my inbox.
- 04
Its boundary is: never [action] without my review.
Example: never send.
- 05
Every [day and time], it sends me [deliverable] so I can verify its work.
Example: every weekday at 7 AM, the brief, to the address I chose.
Those five lines are boxes 1, 1, 3, 5, and 6. Boxes 2 and 4 take longer, because they hold your context and your process, and they are the two that separate a working AI agent from a demo.
What happens after the canvas?
You build. On a no-code platform the canvas maps onto the five moves every AI agent is made of: name it, tell it the job, switch on its abilities, connect your tools, run it. The walkthrough for a first build is how to build your first AI agent, and the free Agent Lab session does one live in an hour. When you get to a second and third AI agent, each one gets its own canvas. One AI agent, one job, one sheet.
Frequently asked questions
What is an AI agent onboarding canvas?
A one-page worksheet you complete before building an AI agent. It has six boxes: the job title and measurable goal, the persona and operating context, the specific systems it can read or write, the step-by-step procedure with the reason for each step, the guardrails on what it cannot do without a person, and the report it delivers for you to review. The finished canvas is the brief you paste into the build platform.
Why fill out a worksheet before building an AI agent?
Because the brief is the product. Opening the platform first tends to produce an AI agent that has access to everything and a job described in one vague sentence. Writing the six boxes first produces an AI agent with one job, scoped access, written limits, and a review routine.
Which boxes take the longest?
Persona and context, and the procedure with the why. The other four are one or two lines each. Those two hold your knowledge of the business and the process, and they are what makes the output sound like your team instead of a generic assistant.
Can I use the canvas for more than one AI agent?
Yes. Each AI agent gets its own canvas, because each one has one job. A team that needs meeting prep, inbox triage, and a weekly report fills out three sheets and builds three AI agents.
What is the first guardrail every AI agent should have?
Draft, never send. The AI agent writes into a drafts folder and a person reads, edits, and sends. Everything else in the guardrails box builds on that one rule.
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 AI agent that is already running.