What has been working well for me lately is having a model like GPT 5.6 Sol do the "core" implementation that requires a lot of effort/thinking...
Then you can come in with Grok 4.5 to add the final touches and help with deployment & DevOps. Works rather well!
@thsottiaux It's significantly better... Allows bursting work in longer durations when needed, not having to worry about being cut off every 5 hours, which is great!
In an ideal world, the limit bar would go down even slower!
@AbhiCodes15 "If the executives or managers of a corporation directed a bunch of employees to build their product, can you still say the company/corporation really built that product?"
This is why GPT 5.6 Sol at high effort/thinking levels *still* can massively benefit from an advisor (i.e. another agent watching what it's doing and keeping it on track).
It was starting to go down a rabbit hole / get into a loop... but the advisor got it back in line.