What the AI team chart shows
The chart is a picture of a plan, not a system that already exists. It shows AI agents for departments the way an ordinary org chart shows people: who sits at the top, who does which job and where decisions get made. The one difference is that next to each role it also shows what the role may do alone and where a person has to say yes.
From top to bottom, the builder draws four layers.
The owner
One named person who sets the limits, reads the log and can switch any agent off. A chart with no owner is a chart nobody answers for.
One agent per department
Each agent has a narrow job inside one department, such as answering common support questions or filing incoming documents.
Approval gates
A gate appears wherever you set a duty to asks a human first. The agent prepares the step and then waits for a yes.
The audit log
Every action of every agent is written to a record a person can read. The bar under the chart stands for that record.
The animation is there to show the flow. Work moves from a department to its agent, stops at a gate when a person must approve, and ends up in the log. Nothing in the picture is measured from your company, and no real agent is running.
How to use the builder
It takes about three minutes and needs no signup. Work through it in this order.
Switch on your departments
Choose from sales, customer support, operations, finance, HR, documents, company knowledge and research. Leave off anything you do not want an agent near yet.
Set each duty
Every department you switch on comes with a short list of duties. Move each one between works alone and asks a human first. Start strict. You can loosen a duty later, once the log has earned it.
Tick the tools
Mark the tools each agent would connect to, for example a calendar, a CRM, a mailbox or a shared drive.
Read the chart
Look at where the gates fall. If a department shows no gate at all, ask yourself whether that is really what you want.
Copy, download or share
Copy the plan as text, download the chart as an image or share it with a colleague. Using the builder also counts towards the free Lab Pass.
Why every agent gets a narrow job
An agent that can do everything is hard to review, hard to test and hard to switch off. A narrow job fixes that. You can say in one sentence what the agent is for, list the few tools it needs and check its log against that sentence.
These are the working rules behind every agent we design, and the chart is drawn around them.
- Least privilege. An agent gets only the tools and records its job needs, and only the permissions it needs inside them. A support agent that reads help articles has no reason to touch the books.
- A dedicated service account. Each agent runs under its own account, never under a person's login. Its actions stay separate in the logs, and you can cut off one agent without disturbing anyone else.
- Approvals on risky steps. Anything that involves money, a promise to a customer or a change to a record of record waits for a person.
- Every action logged. The log says what the agent did, to what and when, so a human can check it afterwards.
- Accounts you own. The agent runs on accounts that belong to your company, so you keep your data and your access if the work with us ever ends.
- An off switch. You can switch an agent off at any time, and the people who rely on it know what to do while it is off.
The gates, the log bar and the one-agent-per-department layout are those rules in picture form. If a plan starts to look crowded, split a department's duties between two agents instead of widening one agent's access.
What usually stays with a human
A good rule is that an agent may prepare and a person decides. These are the kinds of step that usually keep a gate in front of them, whatever the department.
- Money movement. Paying a bill, issuing a refund, sending an invoice or changing a payment detail.
- New commitments to customers. Quotes, discounts, delivery promises, contract wording and any exception to a policy.
- Changes to a record of record. The customer master record, the ledger, the employee file, a signed document. An agent can propose the change and show the difference. A person applies it.
- Anything legal or medical. Advice, interpretation or decisions that carry legal or health consequences belong to a qualified person.
- Decisions about people. Hiring, pay, discipline and leave stay with people. An agent can organize the paperwork around them.
Gates are not a sign of distrust. They are what makes it reasonable to run an agent at all, because a person sees the proposed step before it takes effect. After a good stretch of clean logs you may decide to move one low-risk duty from asks a human first to works alone. Do it one duty at a time, and write down why.
Department examples
These examples show the pattern the presets follow. Your own list will differ, and they are illustrations rather than a recommendation for your company.
| Department | Often works alone | Usually asks first |
|---|---|---|
| Sales | Logs new leads, drafts follow-up notes, keeps pipeline stages tidy | Sends a quote, offers a discount, changes a deal value |
| Customer support | Answers common questions from approved articles, tags and routes tickets | Refunds, exceptions, replies to upset customers, anything about safety |
| Operations | Sends reminders, compiles a daily status, checks lists for gaps | Places orders, changes schedules that affect a crew, switches a vendor |
| Finance | Matches receipts to entries, drafts invoices, flags odd items | Sends an invoice, pays a bill, posts a journal entry |
| HR | Answers policy questions, proposes interview times, drafts job posts | Anything about hiring, pay, discipline or leave |
| Documents | Names, sorts and files incoming documents, pulls out key fields | Deletes or overwrites a record of record, sends a document outside the company |
| Company knowledge | Answers staff questions from approved documents and shows the source | Publishes new policy text, settles a conflict between two documents |
| Research | Collects public information with links and summarizes it | Anything sent to a client or quoted as fact before a person has checked the sources |
If a department looks too risky to start with, begin where the middle column is long and the right column is short. The first AI agent picker helps you compare your processes and choose one to start with.
How an engagement works with us
Once a plan is on screen, the next step is not to build all of it. We start with one agent.
Request
Send the plan you built with a few lines about the work. The request form is on the home page and at the bottom of this page.
Short call
We talk through your tools, your limits and which department to start with. If an agent is not the right answer, we say so.
Written scope
The goal, the tools, the approvals and the success measure are written down before anything is built.
Pilot before rollout
One agent in one department, with approvals on risky steps, every action logged, least-privilege access on its own service account and accounts you own.
Review and extend
You read the log and decide what to keep, change or switch off. More agents follow only if the pilot earned them.
For the advisory side, see AI consulting. For building and connecting the agents, see our AI automation agency page. The proof page shows how we run our own AI agents every day.
When this is not a good fit
The builder is a thinking aid. It will not help much in these cases.
- Nobody can describe the work. If a department's duties live in people's heads, write them down first. The AI readiness audit shows where the gaps are.
- You want a number. The chart shows structure, not savings or return. For the money side, try the AI agent ROI calculator, another free planner.
- Your team is one person in one tool. A chart full of agents would add more structure than the work needs. One agent, or none, may be enough.
- Most of the work is regulated and no rules exist for AI yet. Write the rules before you draw the chart.
- You want a finished design. The chart is not an architecture, a security review or a legal opinion.
In any of these cases a short call is usually a better first step than a chart.
What to prepare before you ask for a scope
- The chart or the copied text plan, so we start from your choices.
- A list of the tools each department uses and who owns each account.
- For each duty you marked works alone, one sentence on what a good result looks like.
- For each gate, the name of the person who would approve.
- A note on sensitive data: health, finance, children, voice recordings, personal files.
- One named owner on your side who has time for a pilot.
No chart yet? Build one now, then send it through the form below. You can also call 773-800-1520 in English or Polish, Monday to Friday from nine to five and Saturday from ten to two, Central time.
Questions answered
What is an AI team builder?
Is the chart a recommendation for my company?
Is anything I choose sent anywhere?
Why one agent per department instead of one agent for everything?
What does asks a human first mean?
Which duties should always keep a human gate?
How would these agents connect to our tools?
Does using the builder count towards the Lab Pass?
Can you build and run the agents on my chart?
Is this legal, security or financial advice?
Scope your first agent
Send us the team plan you built. We reply with a written scope for the first agent, with approvals and logs.