Google-Agent
Google-Agent is the user agent of Google's hosted AI agents when they visit websites and take actions for a person, like filling a form.
Google-Agent is the name Google's AI agents use when they visit websites on someone's behalf. When an agent running on Google's systems browses to a page, reads it or takes an action there because a person asked it to, the request carries the Google-Agent user agent.
It's a sign of where things are heading. Some of your site's visitors are now software doing a job for a person, like comparing prices or booking a table, instead of the person themselves.
How it differs from a crawler
A crawler like Googlebot reads the web on a schedule to build an index. Google-Agent only visits when a person has asked an agent to do something. Google lists it among its user-triggered fetchers, and says that since a user requested the fetch, these fetchers generally ignore robots.txt rules.
So blocking it in robots.txt isn't a reliable way to keep it out. If you need to, you'd do it at your server or firewall, and you'd be turning away people's errands too.
An example
For example, someone asks an agent to find a coffee subscription under £15 a month that delivers to Leeds, and to sign them up. The agent searches, visits Fernhill Coffee's pricing page as Google-Agent, reads the plans and delivery areas, and fills in the sign-up form.
If the price is only in an image, or the delivery areas only load after a script runs, the agent may move on to a site where it can read them.
Checking it's really Google
Anyone can put Google-Agent in a request. Google publishes the IP ranges these agents use, so you can check a visit is genuine.
In AnswerOrbit, the Bot visits tab in Website shows the AI bots that visited your site once you connect your server logs. We count a bot's visit when its address is on the list its company publishes, or looks up to the company's own host name. Requests that fail the check are shown as Unverified requests.