I finally tried Codex today and figured it might be fun to do my first Substack Live so you could see my initial reactions. Overall, it did feel like a step up from Claude Code.
I wanted to test a few specific things:
Does it have good judgment? Does it build on my work to make it better than what I would get if I did everything myself?
Is it good at being delegated to? If I give it an unspecified task, can it figure out how to do it well? Does it “manage up” to me in a helpful way?
Does the UX for creating and managing agents/skills/automations/whatever feel different/better than Claude Code?
To do this, I attempted a few tasks:
Update Couch to 5K for AI to have a Codex path in addition to a Claude Code path
Set up an invoice tracker + agent that creates invoices and manages client status for me
Create a skill that helps me think more ambitiously
Overall I was very impressed and I understand why everyone at the bleeding edge is using Codex over Claude Code now.
Thanks to everyone who joined live! I had a lot of fun and hope to do more of these. What would you be interested in seeing?
HG
