A good part of being good engineer is knowing which library to use, when. LLMs don't take enough time to decide which one is a good library in a particular use case because that layer seems missing unless explicitly prompted to do so.
The strongest competitor to gpt 5.5 is not Opus 4.7, it is @Cursor Composer 2.5 because you get almost unlimited tokens on it through Cursor and it is so competitive.
@ajambrosino The front-end design skills of codex need improvement compared to claude. Make server management or CI-CD or deployment more developer friendly.
@GregKamradt I use it from cursor session with codex extension. It generally stops after 30-40 mins. So I just queue a lot of continue. Have gotten up to total of 2-3 hours with that.
Being a sceptic, I dont let codex by @OpenAI work autonomously for too long, but as I have started using it, I am getting confidence to let it run for this long.