About

Agent Reality Check exists because "autonomous" is easy to claim and hard to verify.

We test AI agents on identical real tasks and report what happened: did it finish, what did it cost, and how often did a person have to step in. The aim is simple, honest numbers that anyone can reproduce.

The project is pre-launch. No results have been published. When they are, we will share prompts, versions, dates, and logs, and disclose any sponsor relationships. Sponsors never change scores.

Founder bio and contact: [placeholder, add before publishing].