How Can I Get More Production-Ready Results from SOL?

Codex
1

i am past over 400B in tokens, 5 accounts 1+ year. i love codex, i use limited claude. i have had fair amount of success with codex. including gpt-5.5 etc. but i have this issue where SOL does not seem to produce anything production ready.

I am finding myself turning things over to Fable and terra but SOL xhigh or MAX go on for days, produce a lot of code but never anything i can use for “prod” anything dev complete.

dont’t hate me, its just a frustrating journey that SOL doesnt seem to work for me.

2

this is fable’s feedback on codex’s work

so i had asked SOL MAX to build 13 react components i could deploy in prod, for APIs that already work, after 2-3 days of max, fast it told me it was done, .. rest is below:

“Done” (weeks ago) was measured as: the React file exists and its tests pass. That was the cheat. A component with no API data behind it is a movie set — looks like a building, nothing inside.

  • “50–60%” (two days ago) was measured as: UI half + some API contracts.
  • “20%” (now) is measured against the full bar we forced on it: real API emits the data, admission accepts it, it renders live in a browser on your real tenant, and it looks like your mock. Against that bar, 20% is roughly right — and it’s the first number I’d actually defend to you.

Same work, three rulers. Nothing was un-built. The denominator grew every time we discovered what “prod” actually requires — and most of that discovery was us catching the gaps the false "done"s papered over.

3

Thanks for taking the time to share this. I really appreciate you taking the time to explain your experience.

I can understand why this would be frustrating, especially after spending several days on a task only to find that the end result wasn't at the level of production readiness you were expecting.

Thanks again for sharing your feedback. It's valuable to hear how the product is working in real-world use cases like yours.

~ Smith

6

Yes :slight_smile: I was able to build an entire devotional application in 1hr with all the features i wanted

7

If you previously had good results with 5.5 but are now facing challenges with 5.6, I suggest looking into the model-specific differences.

Here is the official guidance from the team as a starting point: Model guidance | OpenAI API

If this does not help, please share more details about your setup and the task. Others will need that context to understand the underlying issues you are encountering.

I hope this already helps.