What an AI agent actually costs,
and why quotes vary five times.
The reasoning is cheap and the interface is reusable. What takes weeks is connecting systems that were never designed to talk. Here's where the hours really land.
Quotes for the same AI agent routinely vary by five times, and it's almost never because one vendor is greedy. It's because the thing being priced isn't the agent.
An agent is mostly a renderer with a schedule. The reasoning is cheap, the interface is reusable, and the part that takes weeks is connecting it to systems that were never designed to talk to each other.
Once you know where the hours actually go, you can make two quotes comparable, and you can tell within one question whether the person quoting has built one before.
The one number that drives the price
How many separate systems does it have to read from or write to? Your CRM, your inbox, your calendar, your billing, your project tool, whatever actually holds the work.
Cost scales close to linearly with that count. Everything else, the logic, the layout, the scheduling, is largely fixed and largely reusable between builds.
If someone quotes you a single figure without asking this question, they haven't scoped it. They've guessed, and the overrun will arrive in week three.
Where the hours actually land
Five phases. The middle one is nearly all of it.
Deciding what it should say
What the agent surfaces, what it stays quiet about, what counts as important enough to interrupt you. Fast, and it's the part most people expect to be slow.
Rough share of the build: small, but skipping it produces an agent nobody reads after week two.
Connecting each system
Authentication, permissions, the specific shape of your data, the fields that are named one thing and mean another, the record that three people fill in differently.
This is where essentially all the time goes. Expect it to run several hours per system on a stack the builder already knows, and materially more on one they don't.
Rough share of the build: the overwhelming majority. Price this per system or you aren't comparing quotes, you're comparing optimism.
The part that runs it on a schedule
Something has to wake it up, and something has to notice when it fails. Reusable between builds, so it should barely move the price. If it's a big line item, ask why.
Watch for: a quote where scheduling costs more than a connector. That ratio is backwards.
The interface you look at
Also reusable. A dashboard is a view over a file the agent already produces. Genuinely custom design is a real cost, but most builds don't need it and shouldn't pay for it.
Watch for: a large design line on an internal tool three people will ever open.
Keeping it connected
The line almost nobody quotes. Connected systems change. A field gets renamed, an API version retires, a permission scope narrows, and the agent quietly starts returning less than it did.
Fails when the proposal ends at handover. A build with no maintenance line is a build that degrades on somebody else's schedule.
Why the same agent gets quoted five ways
Two vendors quote the same daily briefing agent. Vendor A comes in low. Vendor B is several times higher and looks unreasonable.
Vendor A has built this on your exact stack before. The connectors already exist, so they're selling you configuration and their number is honest.
Vendor B hasn't. They're quoting original integration work against systems they'll meet for the first time, and their number is also honest.
The quotes aren't comparable and neither vendor is wrong. The question that separates them isn't price. It's "which of these systems have you connected before, and can you show me?"
The trap is that A is only cheaper if your stack really is their stack. One unfamiliar system in the list and the gap closes fast.
Four questions that make quotes comparable
- Price it per connected system. One line per integration, not one figure for the build. This alone exposes most of the variance.
- Which of these have you connected before? Ask for something specific, not a logo wall. Familiarity is the entire difference between the two quotes above.
- What happens when one of them changes? If there's no answer, you're buying a snapshot rather than a system.
- What do I own at the end? The logic, the connections, the credentials. If the answer is access rather than ownership, that's a rental and should be priced like one.
Where to start
Send the system list first
Write down every system it must touch and send that before anyone quotes. It changes the conversation from a number to a scope, which is the only version worth having.
Don't buy an install
A single-system agent is a project. A freelancer will do it for a fraction of what a connected build costs, and you should let them.
Ask who has done it before
At four systems the integration experience matters more than the hourly rate, and it's the only variable that reliably predicts whether it ships.
How we quote it
Per connected system, with the reusable parts named as reusable, and a maintenance line that exists whether or not you take it. If your stack is one we've wired before, you should pay less for that, and you do.
MarqueeOS installs the connected system and hands it over. If your list is one system long, we'll tell you on the call that this isn't an install.
Keep reading
The 6-Point Check For AI Work Before It Leaves Your Company
Most AI failures don't look like errors. Here's how to grade output you didn't build, and turn it into an error rate you can actually act on.
Read the guide → Operations · 6 min readThe 30-Minute AI Inventory: What Is Running In Your Business And Who Owns It
Four questions, four groups, one specific order. Surfaces the undocumented setup carrying real volume before the person running it resigns.
Read the guide →Bring a quote
and we'll read it with you.
Thirty minutes. Bring your system list or someone else's proposal and we'll break it into the lines above, whether or not we end up doing the work.
Book a 30-minute call →