In short
- Functions turn a profile into an actionable contact point.
- v1 can work by email, without publishing private recipients.
- Later, calendars, webhooks, calculators and APIs can connect to the same functions.
Why functions need their own names
An agent can do little with a nonspecific contact form. It needs an action that describes what's wanted: request a rate, request a quote, request a meeting, request a callback or ask a question.
These names also help search: an agent can look specifically for businesses that offer a certain function.
Email-based functions in v1
In the first version a function can be delivered internally by email. The agent sees the function, but not the private recipient address.
Businesses can maintain different internal recipients per function — for example sales for quotes and advisory for callbacks.
The path to real operations
Once a business is more mature, a function can be automated further: calendar booking, webhook, calculator, CRM entry, API call or structured questionnaire.
The advantage: the public agent layer stays equally understandable, while internal execution becomes more professional.
Checklist
- Every function has a clear name
- Every function has a short agent description
- Every function has an internal recipient rule
- Test calls use dry_run or preview_only
- Functions appear in agent.json and the Agent Index
Frequently asked
Which functions should a business activate first?
For most businesses, ask_question, request_quote, request_meeting and request_callback are enough to start. Industry-specific functions come after.
Are functions dangerous?
Not when they're clearly bounded. Private data stays hidden, and the function only describes an allowed request or action.
Ask Google AI
Ask ChatGPT
Ask Claude