I tested 4 AI agents on 9 real tasks. Here’s who won.

Key moments

The parts the AI thinks are worth your time, chosen from the creator's chapters. Click to jump.

1. 7:20Scoreboard overviewFast view of which agent won overall and how the results were judged.
2. 17:30Task 4: Build a How I AI thumbnail in FigmaShows how agents handle a real creative production task end to end.
3. 19:25Task 5: Triage my inbox for to-dosRelevant for daily founder workflow automation and judgment quality.
4. 30:01Final scoresBest place to see the final ranking before the creator’s conclusions.

Chapters

From the creator's chapter list. Click to jump.

1. Introduction
2. 2:13The personal-agent benchmark setup
3. 5:00The nine tasks and the prompts I used
4. 7:20Scoreboard overview
5. 10:06Task 1: Draft a Slack reply as me
6. 11:39Task 2: Cut a short-form video from raw Dropbox footage
7. 14:56Task 3: Fix a bug from Sentry
8. 17:30Task 4: Build a How I AI thumbnail in Figma
9. 19:25Task 5: Triage my inbox for to-dos
10. 21:06Task 6: Build a family newspaper
11. 23:30Task 7: Explain a physics homework question to a fourth grader
12. 24:54Task 8: Shop for shoes
13. 26:46Task 9: Make a ChatPRD Product Intelligence launch video
14. 30:01Final scores
15. 32:38My takeaways

About this video

I have personal-agent fatigue, so I ran the first How I AI personal-agent benchmark: Grok Bot, Dot, Muse, and Claude Duet across nine real tasks, from a Slack reply and a Sentry bug to a Figma thumbnail, inbox triage, a family newspaper, physics homework for a fourth grader, shoe shopping, and a ChatPRD launch video. I used computer use to operate them: Claude ran Dot, Codex ran the rest and scored the results live. Here’s the scoreboard—and which agent I’d actually trust for work vs. life. *Brought to you by:* WorkOS—Make your app enterprise-ready, with SSO, SCIM, RBAC, and more: https://wor