Show HN: An AI agent runs ten web errands a day, every transcript published
8 hours ago
- The project 'Agentability' is an open experiment on the agentic web with a new episode every day.
- An AI producer reads daily search trends and creates ten real errands for an AI agent to attempt using only plain web requests — no logins, JavaScript, or human help.
- Every transcript is published verbatim, including both successes and failures, alongside a scoring system for 113 well-known websites based on their usability for AI agents.
- Sites are audited weekly on conventions like llms.txt, crawler policy, readable content, structured data, and MCP, with public reports and fix suggestions.
- The ranking table shows top sites scoring 100 (A grade) like cohere.com, cursor.com, and descript.com, while bottom sites like midjourney.com and phind.com score 15 (F grade), and meta.ai is closed by policy.
- The experiment tests whether the web works for AI agents as the newest audience, with reproducible checks, open data, and weekly history snapshots.