How Much Do AI Agents Cost?
A first agent role typically runs $1,000 to deploy and $500 a month to operate, which is about $7,000 in the first year. Read that as the entry point for one narrowly scoped role, not as the price of an agent in general. A role that reads from more systems, handles more volume, or carries a heavier approval path costs more, and it is priced before we start rather than after.
What that entry price assumes
It is a real number, and it assumes a specific scope. Five things, and each one you are missing moves the price:
One workflow, not a whole function
One or two systems to read from
A small number of document formats to handle
One approval route above a threshold
A rule set that already exists somewhere, even if it is in somebody's head
If your workflow misses two or more of those, the answer is a scoping conversation rather than a price list, and we will tell you which it is before you commit.
Most published answers to this question either talk about software seats or quote a five figure enterprise programme. Neither one is what a mid-market company is actually buying.
What the price is for, and what it is not
You are not buying a seat. You are not buying tokens. You are buying a role, where the agent reads everything and a person approves the exceptions.
Included | Not included |
|---|---|
The deployment, built for your workflow | Your software licences |
The rules written down so the agent can follow them | The hours your team spends approving |
The agent running continuously | Any system we have to license on your behalf |
The exception queue your team works from | Third party data feeds you may already pay for |
Software stays yours and stays in your name. monday.com is typically billed to you directly at around $50 a seat, so you keep control of the account and the data. That is deliberate. If the engagement ends, nothing of yours leaves with us.
The three costs nobody quotes you
This is the part that decides whether the maths works, and it is missing from almost every answer on the topic.
1. Writing the rulebook down. The agent can only apply a rule that exists somewhere it can read. If your process lives in one person's head, someone has to write it out. Most companies already have the answer spread across old emails, a procedure document nobody updates, and a senior person's memory. Getting it into one place is the real first project, and it is usually smaller than people fear. It is also the only part you keep forever, because it is the same work you would need to onboard a person.
2. The approval time. Nobody's agent approves its own work on day one, and it should not. A person reads the exceptions. On a well scoped role, the queue is short enough that this is less time than the role used to consume, but it is not zero, and any vendor telling you it is zero has not run one in production.
3. The systems count. Every additional system the role has to read from adds work. Email and one portal is a different project from email, three portals, an ERP, and a spreadsheet someone maintains by hand.
Why it is priced per role and not per seat
Per seat pricing rewards the vendor for adding users. Per role pricing rewards the vendor for the role actually working, because we only get paid while the role is running and delivering.
That is also why the comparison to a hire is worth doing carefully. A mid-market operations hire commonly runs $70,000 to $90,000 fully loaded once salary, payroll taxes, benefits, and the management time to run the role are counted. One agent role for the same workflow starts around $7,000 in year one.
The comparison only holds if the work is rule-driven enough for an agent to own it end to end. If the role is mostly judgment, you will pay for both, and that is the most expensive outcome available.
What makes a deployment cost more
Four things, in the order they actually move a quote:
Document variety. Ten invoice formats is more work than one. It is not a blocker, it is a scope.
Approval complexity. One approver above a threshold is simple. A routing matrix by customer, region, and value is not.
Volume. More volume is not automatically more cost, but it changes what the exception rate has to be for the maths to work.
Integration reality. Where an API exists, work is simple. Where it does not, the agent reads the same screens your team reads, which works and costs more to build.
What a first year looks like
Most companies do not deploy five agents at once, and the ones that try usually regret it.
Stage | What happens | Roles running | Monthly total |
|---|---|---|---|
Month 1 | One role deployed, rules written, approval gate set | 1 | $500 |
Months 2 to 3 | One full cycle measured, win rate compared | 1 | $500 |
Months 4 to 6 | Second role deployed on the next worst workflow | 2 | $1,000 |
Months 7 to 12 | Third role, and the first one is now boring | 3 | $1,500 |
Those monthly figures are the entry price of $500 multiplied by the number of roles running. Add up the running months and you get $13,500, plus three deployment fees. A first year with three roles in it runs about $16,500. Compare that against one hire, not against a software subscription, because it is not a software subscription.
What to ask any vendor, including us
Does the agent read every line or does it sample? A sample tells you the size of the problem. Only full coverage recovers anything.
What happens when it is wrong? There should be a specific answer about the exception queue and who reviews what.
Who owns the rules we write? You should. They are your process documentation either way.
What does month one actually cost us in people's time? Anyone who says zero is guessing.
Where this goes next
Cost is the second question. The first one is whether the work is the right shape for an agent at all, and that decision method is here.
Should you hire another person or deploy an AI agent?
How to decide which roles your AI agents should own
AI agents for accounts payable teams
What is workforce design for the AI era?
Frequently asked questions
Is there a minimum number of agents? No. Most companies start with one role, on purpose, because a single role done properly proves the model and produces a number you can take to the next decision.
What does the $1,000 deployment cover? The build for that specific role: reading from your systems, the rules, the exception queue, and the approval gate. It is a one time charge per role, and it is the entry point rather than a fixed rate.
Is $500 a month the price for any agent? No. It is the entry point for one narrowly scoped role, which is the scope we recommend starting with. A role covering several systems, high volume, or a complex approval matrix costs more, and you get that number in writing before anything is built.
Can we cancel? Yes. Nothing is on a long contract. Your software stays in your name and the rules we wrote stay with you, so the work is not lost if you stop.
Why is this so much cheaper than the enterprise products? Because you are not buying a platform licence or a transformation programme. You are buying one role, on your existing systems, with a person still approving the output.
What if the agent does not work out? Then you have spent a deployment fee and a few months of running cost, not a headcount, and you have process documentation you did not have before. That is the honest downside, and it is why we start with the workflow where the return is measurable fast.
