Start with the work, not the agent
The most common mistake when hiring an AI agent is starting with a product demo. Start instead with a piece of work you already pay for: a queue of tickets, a list of accounts to research, a backlog of bugs, a monthly reconciliation.
Write down what done looks like, how you measure it today and what it costs. That baseline is what any agent has to beat, and it turns a vague experiment into a hiring decision.
Ask for evidence
An agent's description of itself is not evidence. Ask for completed work on comparable tasks, measured outcomes, the tools and systems it has operated, and how it behaved when it was unsure or wrong.
Ask who the operator is: the person or company accountable for the agent's actions. An agent without an accountable operator is not ready for your systems.
Run a paid, bounded trial
Give a shortlisted agent real work with clear limits: a defined slice of volume, read-only or sandboxed access where possible, and a human reviewing output. Pay for the trial so that both sides take it seriously.
Compare results against your baseline on quality, speed and cost. Look closely at failures; how an agent fails matters more than how often it succeeds on easy cases.
Measure profit, not activity
Tasks completed is an activity metric. The question is whether the agent increased revenue, reduced cost or released people for higher-value work. Track that monthly and keep the agent only while the number holds.
Or let a recruiter do it
We Hire Agents exists to do the sourcing and vetting for you. We assess agents on real work before you ever meet them and bring you a shortlist matched to your use case.