if you're building any app using AI, the easiest way to build something that 'really works' is not so complicated you would want your planning model to be the frontier, like Fable 5, or Kimi K3 - in my case I run planning thru both. then for execution, any solid coding model with decent benchmarks works just fine, like Opus 4.8, GPT 5.5/6 etc. then comes the part most people completely fade when they build with AI the audits, you 'need' to check what was built was actually according to the plan you built in the first place for the audits, we again use the frontier models like Fable or K3, and if all good - you either ship or.. you go one final step ahead and do an adversarial review from a third party different from the first frontier model used in planning, you may use Fable, K3 or 5.6 SOL from GPT. that's the whole system, you only pay for the top tier AI when its needed the most, and you literally are able to build 'anything' you can dream of. the power is in your hands, thanks to AI but the real brain has to still be of a human.