The AI Agent Buyer's Guide
Every vendor demo looks magical. This guide is the checklist for what happens after the demo: the questions that expose weak offerings, the red flags worth walking away from, and how to run a low-risk pilot.
Start with the job, not the tech
The biggest buying mistake isn't choosing the wrong vendor — it's shopping before you've defined the job. "We should use AI" is not a job. "Every missed call gets a text back within a minute, and new leads get followed up until they book or say no" is a job. Write down two or three jobs like that before you talk to anyone. If you're not sure what's automatable, our first-week automation guide is a good warm-up.
The three ways to buy
Cheapest on paper. The hidden cost is your time: prompt-tuning, integration debugging, and being the person who notices when it silently stops working. Fine if you have a technical person who owns it.
You get exactly what you spec. The risk shows up in month three: who maintains it when an API changes or the workflow breaks? Handover without a maintenance plan is how automations quietly die.
A team builds the automation on infrastructure they operate, monitors it, and fixes it when it breaks. You review results instead of logs. This is the model we run — see our pricing for how we structure it.
Questions that separate real vendors from demos
- "What happens when the agent is wrong?" The honest answer involves approval queues, escalation to a human, and logs you can read. If the answer is "it's very accurate," keep moving.
- "Who owns my data and my accounts?" You should hold your own keys, or be able to revoke access in one step. Be wary of anyone who wants your passwords routed through their staff.
- "What does the agent refuse to do on its own?" Good systems hold anything involving money or new commitments for human approval. Full autonomy over payments is a red flag, not a feature.
- "How do I leave?" Ask what an export looks like and what stops working the day you cancel. Lock-in you discover at cancellation time is the expensive kind.
- "Can I talk to a current customer in my industry?" One real conversation beats every case study PDF.
Red flags
Run a two-week pilot before you commit
- Pick one job from your list — the one with the clearest before/after.
- Agree what "working" means in numbers you already track (response time, booked appointments, invoices chased).
- Insist on approval mode for anything customer-facing during the pilot.
- At the end, compare against the two weeks prior — not against the demo.
For a deeper comparison of agent types and what they're realistically good at, read the capability overview next, or check how agent pricing models compare.
Want this set up for you?
We set up and run AI operations for service businesses — on a private stack we manage, with you approving anything that matters. No software to learn on your end.
Book a Free Discovery Call