I tested 4 AI agents on 9 real tasks. Here’s who won.
Claire Vo of How I AI benchmarks four personal agents, Dot, GrokBot, Muse and Claude Duet, on nine real tasks and concludes GrokBot suits work jobs and Muse suits life jobs.
The gist
-
Vo had Codex, her favorite computer use agent, check that every agent had the same connectors and skills before running the benchmark.
-
Muse was fastest in four of nine tasks and won short-form video, family newspaper and homework help, which Vo says took half a second.
-
GrokBot won overall at 93%, taking thumbnails via the Figma MCP, inbox to-dos in a Grok document, and price-comparison shoe shopping.
-
Dot got closest to the real bug fix and Claude Duet found a second bug, but Vo would code directly in Codex and Claude.
-
Claude Duet came last at 79%, slow and asking permission four times, but made the best launch video using the ElevenLabs MCP.
Claire Vo uses it
"Codex is my favorite computer use agent... one tip I have for you all is I had Codex use browser use to check that all my agents were connected to roughly the same stuff and had access to the same skills."
▶ Play at 3:1397 creators recommend it, including Riley Brown, Dan Shipper, and John Coogan.
See all on the Codex page
Claire Vo recommends it
"I've actually been moving a lot of my ChatGPT, like, googly question-and-answer style chats to Muse because it does such a good job and it's so fast. So, like, sort of my go-to — explain how this works, what is this, look this up."
▶ Play at 24:3117 creators recommend it, including Peter Yang, Riley Brown, and Chris Koerner.
See all on the Muse page
Claire Vo recommends it
"Grock is like 100% the best at using Figma... if you're using Figma, use Grok Bot, man. It is so good."
▶ Play at 19:0220 creators recommend it, including Riley Brown, Peter Yang, and This Week in Startups.
See all on the GrokBot page
Claire Vo uses it
"Our thumbnail templates are actually quite lovely, so it's actually pretty easy to use the Figma MCP and create new thumbnails."
▶ Play at 17:536 creators recommend it, including Blair Metcalf, Ezra Mechaber, and Gui Seiz.
See all on the Figma MCP page
Claire Vo uses it
"Claude absolutely crushed this task. Duet crushed this task... I actually didn't think Claude Duet would do a good job and it did a really good job."
▶ Play at 29:293 creators recommend it, including Riley Brown and Peter Yang.
See all on the Claude Duet page250 creators recommend it, including Riley Brown, Chris Koerner, and This Week in Startups.
See all on the Claude pageWhat they recommend · 6 picks
"Grock is like 100% the best at using Figma... if you're using Figma, use Grok Bot, man. It is so good."
"I've actually been moving a lot of my ChatGPT, like, googly question-and-answer style chats to Muse because it does such a good job and it's so fast. So, like, sort of my go-to ..."
"Codex is my favorite computer use agent... one tip I have for you all is I had Codex use browser use to check that all my agents were connected to roughly the same stuff and had..."
Also mentioned
Ads in this episode
Firecrawl, WorkOS. They're paid, so we keep them out of the picks above.
See what How I AI recommends without being paid →Related episodes
Episodes covering similar topics
More from How I AI