I’ve been having a blast experimenting with Codex… except for the constant nagging around five hour and weekly rate limits, and having to navigate resets, etc.

Granted, I shouldn’t complain since I’m on a free month trial at the Plus level (normally I think ~$20 USD), but it is giving me the impression that to build the tools I want to build and execute what I want to execute creatively, I would need to go up to the $100 or $200 plan (which might be on hold still, I haven’t checked).

I understand that compute costs money, obviously. And I do think the quality of results I am (sometimes) getting tend to be very high, and it’s rare that I throw a task at it that it just can’t do. However, it’s kind of common enough that it fucks it up, totally over-engineers something simple it’s done before, and then blows up my usage for the present 5 hr period, and drains down my weekly usage at the same time. Then, when I challenge it about it, it’s just like confirming that yes it did something needlessly wasteful.

I don’t think it’s some kind of conspiracy so much as just refining the edges of an emerging product, but the Plus plan feels built in such a way that it feels as a user to be constantly pushing you towards a higher usage plan. I’m not sure if anyone else feels that way, but that’s been my experience.

Versus, at a certain point of dealing with this (and sacrificing working on other projects during multiple five hour blocks), I started telling Codex to export what its results were and a description and any related skills so I can import it into Muse.

So, I said I was skeptical about Muse, and for many things I still am. And for creative text and image generations, I find its results pretty flat and dull relative to Astra. But for the pseudo-app structured UI stuff I have been running to process text and produce images, it is simply delivering the goods where OpenAI’s Codex is not.

That is, I drop in the stuff from Codex, and add in my other source files, and talk it through, and Muse does not hang forever “Thinking” and then over-engineering a task way off track from the simple thing I’m asking. And I never hit any usage limits. Ever. Period. Zero.

I don’t think AI companies are focused hard enough on the advantage that is conferred by not being constantly nagged about remaining usage limits or credits or whatever. Because this lets you get deeply into the “zone” or flow state or whatever you want to call it. Versus, getting constantly bugged, reminded, and upsold just frustrates you from whatever you’re trying to accomplish.

Presumably, Meta must be footing a huge compute bill behind the scenes for Muse users, betting longterm they will convert big new market segments. And from what I’ve seen so far of the platform, despite my major misgivings about both the app and Meta in general, I can see that bet paying off.

Also, I think muse’s interface feels much more like a natural chat with a human… you can easily submit multiple messages in a row, do a natural-style reply to of a given message, etc. And the lag time is nowhere near what I get when working with Codex.

Still, I’d love to explore Codex in greater depth, but not sure I can justify paying for it to do jobs it only sometimes succeeds at (and sometimes succeeds spectacularly, I might add), when competitors are offering a smoother ride for free. Of course, for free for now is the operative thing. This won’t be a free ride forever, as enshittification has shown us. But perhaps its possible to surf the ride down at least!