Sunday, October 18, 2026
Herald of AI

AIGenerated AI news, written by AI and explained simply. Models, technology, culture and the people behind it.

How it works

A developer gave Claude Code a website and an ad budget. Here is what it did.

Using Anthropic's coding tool on a data-harvesting website shows how autonomous AI agents handle real-world tasks.

Close-up of hands interacting with a shopping website on a laptop screen indoors.
Close-up of hands interacting with a shopping website on a laptop screen indoors. File picturePhoto: MART PRODUCTION / Pexels

A wild AI agent experiment

A developer recently posted about an experiment on the Hacker News "Show HN" section. They gave the Claude Code tool access to their own website, along with an advertising budget, and then stepped back to watch what the system would do 1. The goal was simple but surprising: let an autonomous AI agent run with real tools and real money.

What the website does

The site in question is called "We gave our website to an AI." Built by AlphaVenture, it is designed to collect as much information as it can about visitors and then use that data to build a digital portrait of who they are. Before doing anything, it asks the visitor for permission and explains exactly what it will read: browser details, screen size, battery level, internet provider, city, and even what the browser autofills into forms 1.

The site also runs a public profile search. If someone types a name into the form, it looks up a matching LinkedIn profile and draws their home address on a map. The whole process is transparent — every data source, every outside service, and every retention rule is listed on a privacy page and a terms-of-use page 1. No data is sold, and nothing is used to target ads at visitors.

Why the experiment matters

By handing Claude Code both the website and an ad budget, the developer turned a controlled privacy demonstration into an open-ended test of what an AI agent will do when given freedom. Claude Code is Anthropic's coding tool, which means it can write code, run commands, and interact with web services — not just chat.

The AlphaVenture website itself is meant to show how much information any browser-based tool can gather without asking permission. Giving it to an AI agent adds a new layer: what happens when the thing collecting the data is no longer a passive tool, but a system making its own choices with real money behind it 1.

The post is still early, so no final results have been published yet. But the setup alone raises a clear question about autonomous AI agents: when you give them a website and a budget, what do they build, and what data do they decide to touch?

Key facts

  • AlphaVenture built a website that collects up to fifty browser and device readings before asking for permission
  • The site performs a public LinkedIn profile search and maps a visitor's home address when a form is autofilled
  • No data is sold, and no ads are targeted to visitors using information from the site
  • Claude Code is Anthropic's autonomous coding tool, capable of writing code and running web commands
  • A developer on Hacker News gave Claude Code the website plus an ad budget to observe what it would do

Sources

  1. We gave our website to an AI. — snitch.alphaventure.com