
Some thoughts after trying Meta's AI agent 'Muse' for about two hours, late to the party.
It proactively checks Google Maps review notifications arriving in Gmail and even drafts the replies. As an AI secretary it outperforms Dots.
It also suggested post ideas based on Instagram and Threads analytics data, but honestly the image-generation quality is mediocre. Whether the posts grow from those suggestions is something I'll investigate next.
Still, Muse seems to come with its own dedicated VM (virtual machine), with an environment that has Codex CLI preinstalled.
It looks most efficient to leave image generation to Codex and run Muse as the commander.
I'll keep digging into practical uses for customer acquisition and business efficiency. I'll share the practical parts in condensed form on YouTube too — members of the web-acquisition × AI master course get the teaching materials first. YouTube is right from my profile ✅
The original post is in Japanese — the Japanese text here is the verbatim original; the English and Chinese are direct translations.
Two hours with Muse: it watches Gmail for Google Maps reviews and drafts the replies — with Codex on image duty and Muse as commander.