Blog Trends
When your customer's AI agent calls: how to prepare your business for AI agents
Personal agents from Meta and others now book, cancel and ask questions for their users. What a business needs before they call: identity, rules, clean facts and a record.
Divine Tech AI 4 min read
To prepare your business for AI agents that act for your customers, decide three things before they show up: how you recognize an agent and the person it represents, what an agent may do with and without the customer signed in, and where a person takes over. Then put your facts, prices and policies in a form a machine can read, and log every agent request the way you log a human one.
What changed recently?
Your customers now have agents of their own, and those agents are starting to call, chat and connect to your systems on the customer's behalf.
- Consumer agents launched at scale. Meta released Muse in the US in September 2026, an agent that can send emails, book travel, fill out forms and make purchases for its user (TechRepublic, September 2026).
- A shared standard was proposed. On October 6, 2026, Sierra and Meta announced the Personal Agent Protocol, an open standard for how personal agents sign in, identify themselves and complete tasks with companies (Sierra). Draft 0.1 is public, and it says any part of it can still change (Personal Agent Protocol spec).
- Machines already call. Since July 2025, Google Search has been able to call local businesses for its users to ask about prices and availability (Search Engine Journal).
In a Gartner survey of 4,879 customers in early 2025, 51% said they would be willing to let a generative AI assistant handle customer service interactions for them (Gartner, June 2025).
What should an agent be allowed to do?
Write this down as policy before the first agent asks. Most businesses already have rules for staff, such as discount limits and refund approvals. Agent callers need the same rules, sorted by how much you know about who is behind the request.
| Who is asking | Allowed | Needs approval | Never |
|---|---|---|---|
| Unverified agent (phone, chat, signed-out session) | Hours, services, published prices, availability, policies | Holding a slot that blocks other customers | Account details, personal data, changes |
| Agent with read access granted by the customer | Order status, upcoming appointments, balances on the customer's own account | Sharing documents or records | Changes of any kind |
| Agent with write access granted by the customer | Booking, rescheduling and cancelling inside your normal policy | Refunds, discounts or exceptions above your limits | Changing payment details or the contact on file |
The "never" column matters most. An agent with write access is still software following instructions, and the person behind it may not have meant what the agent understood. Keep anything that moves money or changes who controls the account behind a step the customer confirms directly.
Why do structured answers beat a better website?
Agents read your business fast and literally, with no sense of what is out of date. If your cancellation policy says one thing on the website and another in the booking confirmation, an agent will quote whichever it found first.
Meta put it bluntly: "A personal agent should not have to interpret a support center article" (quoted in CMSWire). Start with one source of truth for hours, prices, real availability and your cancellation, refund and rescheduling rules. Your own AI agent should answer from the same source.
What should you record for every agent request?
For each request from an agent, log:
- which agent made the request, and whether its identity was verified;
- which customer it acted for, and what access the customer had granted;
- what it asked for and what your system did;
- the rule that applied, and who approved it if approval was needed;
- whether a person took over, and when.
Frequently asked questions
Do I need to support the Personal Agent Protocol now?
Not yet for most businesses. The protocol is at Draft 0.1, and its authors say any part can change. The useful work now is the policy, the single source of facts and the logging, which apply to agents on any route.
Can I stop AI agents from calling my business?
You can limit some of them. Google says businesses can turn off its automated calls in Business Profile settings, under advanced settings, or ask the caller to stop calling (Google Business Profile Help). Other agents have no such switch, so a clear policy for agent callers is the more reliable answer.
Should an AI agent be allowed to change a customer's booking?
Yes, when the customer has granted write access and the change fits your normal policy. Without verified access, the agent should only receive public information.
Next steps
Start with the policy table above. It takes an afternoon, and nothing about it depends on which protocol wins. In Aurora, you choose what it does on its own, what needs a person's approval and what it never does, and every action is logged with the rule that applied. Explore the Aurora platform or read about how Aurora keeps you in control.