What Does an AI Agent Actually Cost to Keep Running?
Not the price of building it — the ongoing bill after launch. It's usage-based, not a flat licence fee, and it scales with a handful of specific things. Here's what drives it up or down, so you can estimate your own instead of trusting a number that'll be outdated by the time you read it.
Model Pricing Changes Faster Than This Page Could
Any specific per-request or per-token number we could quote here would likely be stale within months — providers revise pricing regularly, and the gap between a small task and a complex one can be an order of magnitude. What doesn't go stale is what actually drives that number for your case. That's the rest of this page.
What Actually Moves Your Running Cost
Request Volume
How often the agent actually runs. An agent handling a handful of cases a day costs very little to operate; one handling thousands scales up accordingly, roughly linearly with volume.
Model Choice & Task Complexity
A short lookup-and-reply is cheap to run. A request that needs the agent to read a long document, reason across several steps, or call a more capable model costs more per request — and not every task needs the most powerful model available.
Infrastructure & Support
Hosting for whatever glue code and logging sits around the model call, plus a support retainer for monitoring, fixes and adjustments as your connected systems change shape over time.
A Rough Way to Size It Yourself
Before a scoping call, you can get a directional sense of your own running cost with three questions.
Count the Volume
How many times a day, realistically, would this request type happen? Pull it from ticket counts, form submissions or call logs — not a guess.
Rate the Complexity
Is it a short lookup, or does it involve reading long documents and multiple decision steps? Complexity, not just volume, moves the per-request cost.
Multiply, Loosely
Volume times complexity gives you an order of magnitude, not a bill. It's enough to tell whether you're looking at a small monthly number or one worth budgeting for properly.
How We Design Against This Up Front
Right-sized model per step
Not every step in a request needs the most capable model available — a simpler step can often use a cheaper one without a quality drop.
Trim what the agent actually reads
Passing only the relevant slice of a document or record, instead of everything available, keeps each request lean.
Usage monitoring from day one
So a spike in volume or an unexpected retry loop shows up as an alert, not a surprise on next month's invoice.
Running-Cost Questions
Why won't you just give a monthly dollar figure?
Because any single number would be wrong for most readers within a few months, in either direction. Model pricing changes, your own volume changes, and the complexity of what the agent does changes what it costs to run. The drivers on this page stay accurate; a specific number wouldn't.
Does running cost scale with the number of users, or the number of requests?
Requests, not seats. An agent handling ten requests a day for a hundred employees usually costs less to run than one handling ten thousand requests a day for five employees. Volume of work is what drives usage cost, not headcount.
Can running costs spike unexpectedly?
They can, usually when request volume jumps — a busy season, a product launch, a bug that causes retries. Sensible builds include basic usage monitoring and sane limits so a spike is visible and controllable rather than a surprise on an invoice weeks later.
Is a cheaper model always the cheaper choice overall?
Not always. A cheaper model that gets more cases wrong and escalates more often can cost more in support time and rework than a pricier model that handles the same volume cleanly. We weigh this during scoping rather than defaulting to whatever has the lowest sticker price.
How does this relate to what's on the pricing page?
The pricing page covers what we charge to build and support your agent. This page covers the separate, ongoing usage bill you'll pay directly to whichever model or infrastructure provider the agent runs on — a cost that exists no matter who built the agent.
Get a Real Estimate for Your Volume
Tell us roughly how often the request happens today. We'll walk through what that means for running cost, plainly, before you commit to anything.