@simonw Simon, as someone who uses AI & is good at writing, are you refining your wording at all with AI? I feel like I have to send my drafts to AI for approval haha
@thsottiaux Thank you for all you’ve done. I’ve always backed OpenAI but I’m really enjoying both claude code & cursor. Once the new limits kick in I’m cancelling my $200 plan.
@YunfengB@OpenAIDevs as a long time codex user, I am already 80% switched to claude since they nerfed Astra. I give them the same review take and Astra max comes back with no findings and even sonnet 5.5 comes back with a list. Then Astra says “oh yeah I agree with all of that” 🥲
@dmtrinh@housecor Think of a developer who is deeply abstracted away from the customers and then asked to work in isolation, and all the other developers are too busy working in isolation to work together. The pressure to deliver overtakes the capacity for human interaction.
Thoughts on my current flow?
Grok 4.7 finds relevant code (cheap exploration)
Opus 5.5 builds a plan (I just like the plans)
Grok & Astra interrogate the plan (the plan is where it all falls apart)
Grok writes the code (cheap)
All 3 models review the code in 3 stages:
Behaviour
Security
Maintainability
Zero production bugs so far 🤞🏼
FYI, I do all the product thinking up front. Open questions or design decisions are surfaced to me before the models implement.