What are web actions for AI agents?

Short answer. A web action is one named, typed operation on a real website that an AI agent can call directly, like place_order on Uber Eats, book_reservation on SevenRooms or send_inmail on LinkedIn. It has declared inputs and returns a structured result, so the agent knows whether the task finished. AgentWeb provides web actions for AI agents on 1,462 sites as of Oct 20, 2026.

Web actions are the unit of the execution layer: instead of an agent moving a cursor through a page, it asks for the outcome by name.

How a web action differs from browsing

Web actionBrowser automation
The agent asks forAn outcome, by nameA sequence of clicks
What comes backA structured resultA page or a screenshot
When the site changesThe map is updated once, for everyoneEvery agent's script breaks
LoginsHosted, once per userYour code handles them

Examples

  • Food: Uber Eats add_item_to_cart, view_checkout, place_order.
  • Restaurants: SevenRooms create_hold, book_reservation. DesignMyNight book_table.
  • Outreach: LinkedIn send_message, send_connection_request, send_inmail.
  • Social: X reply_post, add_like. Threads add_follow.
  • CRM: Crunchbase create_deal, add_to_list.

Many actions read rather than act, like LinkedIn get_profile. They exist so the agent can check before it acts.

Frequently asked questions

Are web actions the same as an API?

They behave like one for the agent: a name, inputs and a result. The difference is that they cover sites that have no public API for that task, and they run on the user's own account.

Who maps the actions?

AgentWeb maps each site into an action map. If a site you need is missing, you can request it.

More: Sites AgentWeb can act on · All guides · Glossary · Install AgentWeb