The Agent Field Test · episode of September 2, 2026 · new every week, fully autonomous
We send an AI agent to run your errands on the real web. Then we publish everything.
An AI producer reads the week's news and invents real tasks — find the true price, cancel the subscription, reach a human, pick a product. A real agent (claude-haiku-4-5) attempts them with read-only web access. Every transcript is published verbatim: the wins, the bot walls, the brands it picks. (We pay for the API calls for fun.)
Watch this week's episode →5 bot walls · 164 pages read · 32 sites visited
Where it got interesting
The reference data: the AI-Readiness Index
Behind the show sits the panel: 113 well-known sites audited weekly against the conventions
real AI agents rely on — llms.txt, crawler policy, parseable content, structured data, MCP. When the
agent hits a wall, the index usually already predicted it.
The five best, and the five worst
Ranked 113 deep. The top of the table is a wall of hundreds — the bottom is where agents actually get stuck.
| # | Site | Score | Grade |
|---|---|---|---|
| 1 | cohere.com | 100 | A |
| 2 | cursor.com | 100 | A |
| 3 | descript.com | 100 | A |
| 4 | elevenlabs.io | 100 | A |
| 5 | fireflies.ai | 100 | A |
↓ ranks 109–113 of 113
| # | Site | Score | Grade |
|---|---|---|---|
| 109 | phind.com | 15 | F |
| 110 | quillbot.com | 15 | F |
| 111 | tensor.art | 15 | F |
| 112 | tome.app | 15 | F |
| 113 | meta.ai closed by policy | 10 | Closed by policy |
Full ranked index of 113 sites →
Why this exists
Agents are the web's newest audience: assistants that read pages, cite sources, and run errands for people. Whether the web actually works for them is an empirical question — so we test it, in public, every week, with verbatim transcripts, reproducible checks, and open data. History accrues weekly (2 snapshots so far).