AI agent quotes range from a few thousand dollars to six figures. Here's what actually drives the price, what it costs to run one, and how to budget.
Ask five vendors what an AI agent costs and you'll get five answers, some of them a hundred times apart. That's not because anyone is hiding the ball. The price depends on a handful of decisions, and most quotes don't spell out which ones they assumed.
Here's how we think about it.
What you're actually paying for
Every AI agent has two price tags: what it costs to build, and what it costs to run.
- Building it: scoping the workflow, writing the instructions and tools the agent uses, connecting it to your systems, testing it against real examples, and deploying it
- Running it: model usage, hosting, and someone keeping an eye on it after launch
A quote that only covers the first part isn't the whole number.
The five things that move the price
- 1.Integrations: this is the biggest driver. Every system the agent touches, like your CRM, inbox, calendar, or ERP, needs authentication, data mapping, and error handling. One integration is simple. Six is a project.
- 2.Autonomy: an agent that drafts work for a person to approve is cheaper than one that acts on its own. Acting alone means guardrails, approval rules, and an audit trail.
- 3.Data readiness: clean, well-organized documents make an agent faster to build and more accurate. Scattered spreadsheets and tribal knowledge add time.
- 4.Volume and speed: ten tasks a day and ten thousand tasks a day are different engineering problems, and so are "by tomorrow" and "in under a second."
- 5.Risk and compliance: healthcare, finance, and legal work need extra review, logging, and data handling. That's time well spent, but it's time.
Three rough tiers
A single-task agent
One job, done well. For example, an agent that answers inbound leads within minutes, asks qualifying questions, and books a call. One or two integrations, a person in the loop for edge cases. These are the least expensive to build and usually ship in a few weeks.
A multi-step workflow agent
An agent that researches, decides, and acts across several systems: it reads an incoming request, pulls account history, drafts a response, updates the CRM, and flags anything unusual for review. These cost more, mostly because of the extra integrations and review steps.
A team of agents
Several agents with different roles working together, plus a dashboard to watch them, approval controls, and alerts. This is the most expensive tier, and it's rarely where you should start.
What it costs to run
Once an agent is live, three ongoing costs show up:
- Model usage: AI models bill by the amount of text they read and write. For most business workflows that's a small amount per task, but it scales with volume, so measure cost per completed task, not per request.
- Hosting and integrations: usually modest, and often covered by tools you already pay for.
- Monitoring and upkeep: models change, the APIs you connect to change, and so do your processes. Somebody needs to watch quality and cost after launch. That's what our AI Operator service is for.
How to keep the cost down
- Start with one workflow that has a clear before and after, like hours spent per week or time to first response
- Use the tools you already have instead of migrating to new ones first
- Launch with a person approving the agent's work, then widen its autonomy as it proves itself
- Ask for a fixed scope so the number can't drift
Questions to ask before you sign
- Is this quote fixed-scope, or billed by the hour?
- What happens after launch, and what does support cost?
- Who owns the code, prompts, and data when we're done?
- What will this cost per month to run at our volume?
- How will we know it's working?
If a vendor can't answer those plainly, keep shopping.
The bottom line
The cheapest agent is the one aimed at a workflow that's eating real hours every week, built with only the integrations it needs, and launched with a person in the loop. Get that right and the budget conversation gets a lot easier.
Every project we take on starts with a free scoping call, and you get a transparent estimate before you commit. Start a project when you're ready.