In a demonstration of agentic coding capabilities, the team behind AlphaVenture tasked Claude Code with building 'Snitch,' a website that profiles its visitors. Posted to Hacker News on October 15, 2026, the project showcases an AI-generated interface that emphasizes explicit user consent and data transparency rather than automated ad bidding. The site is designed to be 'Best on a computer,' with a clear directive that 'Nothing is read until you choose,' placing user agency at the center of the technical architecture.

Static Logic, Not Autonomous Management

Contrary to claims of autonomous real-time decision-making, Claude Code generated the static code and logic for the site’s data collection routines. The AI wrote the rules for how the browser reads approximately fifty distinct attributes from the user’s machine, including screen resolution, graphics memory, battery status, and operating system details. These readings occur without permission prompts, but the flow is deterministic based on user actions, such as scrolling speed or clicks, rather than being managed by a live AI agent optimizing for revenue.

Functional Integrations, Not Ad Networks

The site integrates with external services like Google Maps, LinkedIn, and weather providers, but these are standard web integrations coded by the AI, not dynamic ad-tech pipelines. For example, if a user’s browser autofills a form, the generated logic triggers a public LinkedIn profile search and maps the home address via Google Maps. The source explicitly states that 'Nothing is sold, and nothing is used to advertise to you,' clarifying that data sharing is strictly for functional purposes, such as city lookup or device identification, rather than for advertising targeting.

Privacy-First Infrastructure

Claude Code implemented a privacy policy that is unusually explicit for an AI-generated site. Data is retained on AlphaVenture’s servers for up to 90 days, tied to the specific machine, allowing a second visit to resume where the first left off. The AI coded a feature that allows users to erase their data completely, removing the data itself rather than just revoking permission. Google Analytics is used strictly 'to see what the site does, not who you are,' reinforcing the non-commercial nature of the infrastructure. The site also includes a 'Rather not be read?' option, allowing users to view what the AI has done without submitting their own data.

Key Takeaways

  • Claude Code generated the site’s structure and privacy text, emphasizing a 'no selling, no advertising' policy.
  • Data collection is based on deterministic user actions (e.g., autofill triggering a lookup), not autonomous AI decision-making.
  • The site highlights user consent with the promise that 'Nothing is read until you choose' and is optimized for desktop use.
  • External integrations (LinkedIn, Google Maps) are functional utilities coded by the AI, not ad-network integrations.

The Bottom Line

This experiment proves that AI can autonomously generate not just the visual interface, but the complex legal and logical scaffolding required for a privacy-conscious web application. By removing the incentive to sell data, AlphaVenture and Claude Code demonstrate that transparency and user consent can be coded as core product features rather than afterthoughts.