An AI agent does not have one price. It has four cost layers: a one time setup fee scoped to your workflows and integrations, a recurring subscription that covers the environment and the people operating it, metered model and third party usage, and your own team's time. VantaSoft's managed plans start at $200 per month, with setup scoped during discovery.
That answer is longer than most buyers want, and it is the reason budgeting for an agent goes wrong. Vendors publish the layer that is easiest to publish, which is the monthly number. The layers that actually decide whether the project lands inside budget are the other three.
This is a breakdown of all four, what moves each one up or down, and a worksheet you can take to any vendor before you commit a dollar.
The four things you are actually paying for
| Cost layer | What it covers | How it is billed | What drives it up |
|---|---|---|---|
| Setup | Discovery, workflow design, integration configuration, environment provisioning, internal QA, and support through your testing period | One time, scoped per customer | More workflows, more integrations, systems without a documented API, messy source data, complex approval rules |
| Subscription | The managed environment the agents run in, monitoring, routine software and compatibility updates, issue investigation, support, and minor in scope configuration changes | Recurring monthly, by agent capacity | More agents in production, dedicated support arrangements, service level commitments |
| Model and third party usage | Language model API calls and any outside subscription the agent needs to do its job | Metered by volume, or a separate vendor subscription | Higher run frequency, longer documents, larger models, more reasoning steps per run |
| Your team's time | Discovery sessions, granting access, testing, approving output, and reviewing exceptions after launch | Not invoiced, but real | Unclear workflow ownership, slow access approvals, no named reviewer |
Only the second layer looks like a subscription. The other three are where budgets break, and they are the three that get left out of the spreadsheet.
Why setup is scoped instead of listed
Almost no serious vendor publishes a setup price, and buyers reasonably read that as evasion. It usually is not. Setup is the part of the work that is different for every customer, because it is the part that touches your systems.
Configuring an agent to read one shared inbox and draft replies for review is a different job from wiring the same agent into a CRM, a project board, and an accounting file with three approval gates. Same product, very different amount of work. A published setup price would have to be either high enough to cover the second case or low enough to be wrong about it.
What you should expect instead is a setup number that arrives after a scoping conversation and before you sign anything. If a vendor wants a deposit before it can describe your workflows back to you, that is the problem worth reacting to, not the absence of a price list.
How to make setup smaller. Start with one workflow rather than a department. Prefer systems that already have a documented API. Decide who approves what before the build starts. Clean the data the agent has to read. Every one of those is a lever you control.
The cost drivers that actually move the number
When two quotes for the same idea come back far apart, it is almost always one of these six differences, not a difference in margin.
- How many workflows. One well defined workflow is a project. Five is a program. Cost tracks workflows far more closely than it tracks agents.
- How many integrations, and what kind. A system with a documented API and clean permissions is straightforward. A system with no API, or one that requires a custom broker, is a build.
- How much judgment the work needs. Classifying and routing is cheaper than work that requires reconciling conflicting sources or weighing exceptions.
- Volume and frequency. Volume mostly shows up in the usage layer, not setup. A workflow that runs twice a day and one that runs two thousand times a day cost about the same to configure and very different amounts to run.
- Approval and oversight design. Every action that must pause for a person is a rule someone has to define, build, and test. This is money well spent, and it is still money. See what access an AI agent should have to your business systems for how to scope that without over granting.
- Data readiness. If the records the agent reads are inconsistent, someone pays to make them consistent. That someone is either the vendor during setup or your team afterwards.
If you are not yet sure which workflow belongs in the first scope, this guide to where AI fits into existing workflows and the eight practical agent roles are the faster place to start. Pricing a workflow you have not chosen is guesswork.
Model usage: the line item people forget
Language models are billed by the token, quoted per million tokens, and split between what goes in and what comes out. Output costs several times more than input at every major vendor.
As of August 2026, Anthropic's published list price for Claude Sonnet 5 is $2 per million input tokens and $10 per million output tokens. OpenAI's published list price for gpt-5.6-terra is $2.00 per million short context input tokens and $12.00 per million short context output tokens. Cheaper and more expensive models sit on either side of those.
A worked example, with the assumptions stated. Suppose one run of your workflow reads about 20,000 tokens of context and produces about 2,000 tokens of output. At Claude Sonnet 5 list prices that is $0.04 of input plus $0.02 of output, so roughly $0.06 per run. A thousand runs a month lands near $60.
Two honest caveats. First, a real agent workflow rarely makes one model call per run. It may plan, call a tool, read the result, and decide again, so multiply by the number of steps rather than assuming one. Second, list prices change, so treat any number in an article as a starting point and check the vendor page before you budget.
The practical point is not the exact figure. It is that this layer scales with volume while setup does not, and a vendor who cannot tell you which side of that line a cost falls on has not thought about your workflow.
Ask one more question here: who pays for it. Any third party subscription or usage cost should be identified during discovery, named in your Order Form, and never charged without your written approval.
What the same thing costs to build in house
The in house comparison is worth doing, and it is usually done wrong, because people compare a build estimate against a subscription. The build is the cheap half.
Start with the labor. In the May 2025 Occupational Employment and Wage Statistics from the U.S. Bureau of Labor Statistics, the national annual median wage for software developers, occupation code 15-1252, was $135,980, and the national annual mean was $148,100. In California the annual median for the same occupation was $174,410. The national median works out to about $11,300 a month in base pay alone, before payroll taxes, benefits, equipment, or the cost of hiring.
Now add the part that does not end. Somebody has to monitor the agent, patch it, notice when a vendor changes an API, retest the workflow after that change, and be reachable when it stops working during your busiest week. That is not a project line. It is a standing responsibility, and it is the reason internal automations quietly stop being used about a year after the person who built them changes roles.
None of this means building is wrong. If you already have engineering capacity, clear technical ownership, and an appetite to operate the thing, in house can be the right call. Our build versus buy framework walks through that decision properly. The mistake is choosing to build without deciding who operates it, because the real choice is not managed versus in house. It is managed versus unowned.
What VantaSoft charges
We publish the recurring side in full. Plans are priced by how many managed agents you run in production, not per seat and not per message.
| Plan | Managed agents | Monthly | Typical shape |
|---|---|---|---|
| Launch | Up to 3 | $200 | A first workflow or two, one owner, a small set of approved systems |
| Team | Up to 6 | $375 | Several roles running across more than one function |
| Scale | Up to 12 | $700 | Agents operating across multiple departments |
| Enterprise | More than 12 | From $1,500 | Capacity and terms set in a customer specific Order Form |
Every plan includes a managed cloud environment isolated per customer, monitoring and routine software and compatibility updates, issue investigation and email support during Pacific Time business hours, minor in scope configuration changes, and human review rules for consequential actions.
A change order applies to major new features or workflows, integrations not listed in your Order Form, large data migrations or cleanup, custom software development, capacity above your tier, and dedicated support arrangements. Those are quoted separately rather than absorbed quietly.
Discovery comes first. A one-time setup fee based on the scope of your agents, workflows, and integrations applies. Setup and monthly billing begin after you approve your Order Form. Enterprise pricing is customer-specific.
Current plan details always live on the Agent Service pricing page, and the agent directory shows the roles those plans cover.
The first agent budget worksheet
Fill in these eight lines before you talk to anyone. They are the inputs any competent vendor needs in order to price your work, and they are the fastest way to tell a real quote from a guess.
| Line | What to write down | Why it changes the price | Cost layer it hits |
|---|---|---|---|
| 1 | The workflow, named in one sentence, and the person who owns it | Unowned workflows expand during the build | Setup, your time |
| 2 | What starts the work, and how many times a month it happens | Frequency drives usage, not setup | Usage |
| 3 | Systems the agent must read from | Each source is a separate integration decision | Setup |
| 4 | Systems the agent must write to or act in | Write access needs more design, testing, and review | Setup |
| 5 | Actions that must stop for a human, and who that human is | Each approval gate is a rule to build and test | Setup, your time |
| 6 | Outside subscriptions or licenses the agent will need | These sit outside the monthly plan | Usage |
| 7 | Internal hours for discovery, granting access, and testing | The most commonly omitted cost in the whole exercise | Your time |
| 8 | How many agents you expect in production after 6 months | This selects your capacity tier | Subscription |
Then apply the test. Hand those eight lines to a vendor and ask which lines change the setup fee, which change the monthly fee, which change usage, and which are on you. A vendor who can answer line by line has scoped work like yours before. A vendor who answers with a single number has not.
Deciding whether it paid off is a separate exercise, and it is worth setting up before you start rather than after. Our guide to measuring the ROI of a software project covers how to define the baseline and the measurement window.
Frequently asked questions
How much does an AI agent cost per month?
For a managed service, the monthly fee is usually set by how many agents you run rather than by usage. VantaSoft's plans run from $200 a month for up to three managed agents to $1,500 and above for more than twelve. Model usage and any outside subscriptions are separate from that monthly fee.
Why will nobody give me a setup price up front?
Because setup is the part that depends on your systems. The number of workflows, the integrations involved, the state of your data, and how many actions need human approval can move the work by an order of magnitude. Expect a scoped number after a discovery conversation and before you sign, not a list price on a web page.
Is there a free trial or a self serve plan?
Not for VantaSoft's Agent Service. It starts with a discovery conversation, then a customer specific Order Form covering your agents, workflows, integrations, and tier. Billing begins after you approve that Order Form. You do get a testing period before you approve regular use.
Do we pay for the language model separately?
Model and third party usage is its own layer, and it should be identified during discovery rather than discovered on an invoice. Any third party subscription or pass through cost should be named in your Order Form and approved by you in writing before it is charged.
What happens to the cost when we add a second workflow later?
Adding an agent inside your current capacity does not change the monthly fee. Going above your tier moves you to the next plan. Major new workflows and integrations that were not in your Order Form are scoped as a change order, so the increase is quoted before the work starts.
Is a cheaper agent a worse agent?
Not necessarily, but check what the price excludes. A low monthly fee attached to a platform you configure yourself is not the same product as a service where somebody else monitors it, updates it, and answers when it breaks. Compare the four layers, not the headline number.
Price the workflow, not the software
The useful question is not what an AI agent costs. It is what this workflow costs to run today, what it would cost to run with an agent, and who is on the hook when it stops working. Answer those three and the budget writes itself.
If you want the scoped version for your own workflows, start with a workflow discovery session. It takes 30 minutes, it produces the inputs above, and it happens before any commitment.
