How can an AI agent look something up online, like a client's website or a property estimate, and save it to the contact?

Use a Sub Agent step in a workflow. A sub agent is a small AI agent inside the workflow that can do a task, use tools and return structured output.

1. Start the workflow with a trigger that carries the input, for example a form submission with the client's website or a property address.
2. Add a Sub Agent step and write instructions, for example: crawl this website and summarize the business, or search online for the estimated value of this address. Use the variable picker to insert the website or address from the form.
3. Give the step the tools it needs. Agents have no tools by default. The main web tools are web search and web extract. A third, web research, takes longer: it researches across multiple sites and pages and returns a complete report, for example a company background before a meeting.
4. Define the structured outputs you want, for example a business summary, name, email, business hours, whether the site has a form, or an estimate and the first image. Structured outputs can be used anywhere later in the flow, such as in an email ("Your estimate is ...").
5. To save a result on the contact, create a custom field first (for example Business summary or Zestimate), then use the Update Contact step to put the output into that field.

If you know the exact page, try web extract on the page URL first. Some sites block crawlers, so you may need another route such as an API (which can cost money) or a web search. A stronger model may do better on harder pages.