Plan Low Implement High

I always see people talking about having some sort of high reasoning orchestrator that delegates tasks to subagents. I feel like this could be counterintuitive at times, multi agent delegation introduces an integration problem when actually producing code. I could see it add value for independent review though. What I do is I plan with Sol 6.1 (absolute work horse, infinite tokens lol), he has all the context needed to create a MD hand off. I go back and forth with him, planning and making sure we’re on the same page, defining - architectural goal - existing dependencies, -constraints - expected behavior Then I give it to astra on High/XHigh to implement. The handoffs always state that they provide the scope of the work, as well as necessary safeguards, but Astra has the freedom to decide on technical implementation. This has worked well for me, what do you guys think? Maybe the approach is shit but im all ears. My primary concern is quality, not necessarily saving tokens since I have pro.