When people search "hire an AI agent," they rarely mean an HR process. They want to know three things: what it takes to get an agent doing a real job, how vendors charge for that work in US dollars, and what makes the bill go up or down. This guide answers all three, with real list prices where vendors publish them and honest gaps where they don't.
A quick disambiguation: this page is from zamp.ai, the AI employee platform. It isn't Zamp HR, a payroll and PEO product, and it isn't zamp.com, a US sales-tax compliance company. Same name, different businesses.
What hiring an AI agent actually involves
The sequence looks about the same whichever vendor you pick, because the problem is the one you face with any new team member: get them productive on a real job without breaking anything.
- Scope the job. Name the outcome, the systems it touches, and where the agent decides alone versus when it escalates to a person.
- Connect the systems. Authenticate the agent against your ERP, CRM, ticketing tool, or inbox, and review exactly what access it gets.
- Set the guardrails. Write down tolerances, approval thresholds, and escalation rules before the agent touches production data.
- Run a scoped pilot. Use a defined segment, real data, and a success condition agreed in advance.
- Expand deliberately. Widen scope once the pilot has earned trust, and feed reviewed corrections into the next rollout.
Most pilots stall on the deployment work, not on the model: data access, governance, and the ability to show a queryable record of what the agent did and why. If a platform can't produce that record, it won't clear an enterprise security review at any price. For a fuller picture of how AI employees fit into operations, see the complete guide to AI employees.
How AI agent pricing works in the US market
Four pricing models dominate. The examples below are other vendors' published US list prices, not Zamp's, and they're here so you know what to ask any vendor.
Per-outcome pricing
You pay only when the agent finishes a defined result. Intercom prices its Fin AI Agent at $0.99 per outcome, where an outcome is a confirmed resolution, a conversation where the customer asks for no further help, or a completed workflow, charged once per conversation (Intercom pricing). The vendor carries the risk of failure, so the per-unit price usually reflects that. Check the contract's definition of an outcome, because that definition is the price.
Per-task or per-action pricing
You pay for each unit of work the agent attempts, whether or not it succeeds. Salesforce sells Agentforce actions through Flex Credits at $500 per 100,000 credits, with a standard action consuming 20 credits, which works out to $0.10 per action (Salesforce Agentforce pricing). Per-task pricing is easy to forecast when volume is steady and hard to forecast when one request fans out into many actions.
Usage-based pricing
Cost tracks the compute the agent consumes: tokens, model calls, or a metered unit. Zamp's own meter is the Agent Compute Unit (ACU), which measures the work an agent does rather than charging per seat. Usage pricing fits work whose volume swings, but you need visibility into what drives consumption, such as task complexity, number of steps, and run frequency.
Per-seat and hybrid pricing
Per-seat pricing is a flat monthly fee per named user, carried over from SaaS. It's losing ground for agents that work on their own rather than assisting one person. Hybrid pricing pairs a platform or seat fee with a usage layer. Intercom's helpdesk plans are an example: seats from $29 to $132 per seat per month on annual billing, plus $0.99 per Fin outcome (Intercom pricing).
The useful question for any vendor isn't "what's the rate?" It's "which unit am I paying for, who carries the cost of a failed attempt, and what happens to my bill if volume doubles next quarter?"
Build vs. buy: the cost comparison
Our guide to building an AI agent in-house breaks down typical build costs: a single production-grade agent generally runs $25,000 to $200,000 and takes two to six months, depending on integrations and regulatory overhead. A multi-agent suite covering several workflows typically runs $250,000 to over $1 million in year one, mostly in integration and evaluation work rather than model costs.
Buying trades that months-long build cycle for an integration cycle measured in weeks, at the cost of some customization depth. Build costs are fairly stable because engineering time is engineering time. Buy-side costs vary with the job. The build vs. buy guide covers the decision framework, and the enterprise AI employee platform comparison lays out how the main platforms differ.
What drives deployment cost, whatever the vendor
Set the pricing page aside and five factors decide what a deployment costs:
- Systems integrated. Each ERP, CRM, ticketing tool, or legacy app without an API adds integration and testing work.
- Volume. A hundred cases a month and a hundred thousand cost differently to build and to run, and the gap is widest under per-task pricing.
- Exception complexity. A few predictable exceptions are cheaper to govern than a long tail of edge cases that each need a rule.
- Compliance and audit requirements. Regulated work in banking, healthcare, or employment decisions needs a full decision trail and sometimes a specific deployment model, such as on-premises or bring-your-own-cloud.
- Implementation support. How much scoping, testing, and rollout your team can carry versus the vendor.
These factors explain most of the price spread between quotes for what looks like the same agent on paper. A $0.99 outcome price tells you little until you know how many outcomes your volume produces and how many exceptions fall outside the definition.
How Zamp prices AI employees
Zamp meters work in ACUs rather than seats, and it doesn't publish a self-serve rate card. Two companies asking for the same role, say an AI employee for accounts payable, can differ on every driver above. One might process 5,000 invoices a month across two ERPs with no regulatory overhead. Another might process 50,000 across four systems with SOX controls attached. A single published number would mislead one of them.
So the quote starts with a scoping call covering the job, the systems, your volume, and your compliance requirements. For context on where AI employees are being deployed and governed today, see the State of AI Employees 2026 report.
Get a scoped quote
If you have a specific job in mind, such as an AP exception queue, KYC onboarding, or an HR ticket backlog, talk to the Zamp team and we'll scope it against your volume and systems.
