About

Can Agents Work? is a public-goods project. It collects the best public evidence about what AI agents can do at work, maps it onto the job market, and gives visible credit to the institutions that measured it.

Who runs it

Eddie Wang runs Can Agents Work as an independent, personal project. X: @eddiewang. GitHub: github.com/eddiewang.

Disclosures

Eddie Wang is the CTO of Tenki. Tenki builds sandboxes for AI agents, GitHub Actions runners, and an AI code reviewer, so it works in some areas that the atlas covers, such as AI code review. No Tenki product appears in the atlas results. Eddie has no other employment, investment, or advisory ties to AI labs, agent companies, or benchmark makers.

The site takes no money from the institutions it covers. Names identify sources and do not imply endorsement or partnership.

Corrections

If a number, summary, or mapping is wrong, send the page link and the correct source to hello@canagentswork.com. We aim to fix errors within 48 hours and record the change.

Suggest a benchmark

We add benchmarks that measure agents on real work and publish their method and results. Send the paper or site, the institution, and the job family it covers to hello@canagentswork.com.

Principles

1. Credit first. Each number names its source, larger than our own name.
2. Primary sources first. Aggregators are for discovery only.
3. No number without a source. The build rejects it.
4. Tasks are not jobs. We show task-level and project-level evidence apart, and say what we do not know.
5. No logos without permission.