I use it for both — side projects and my day job in embedded systems. The key is where the tokens go. More tokens spent on planning, design, spec validation,

I use it for both — side projects and my day job in embedded systems. The key is where the tokens go. More tokens spent on planning, design, spec validation, test generation, and multi-agent review than on writing the actual code. The review pipeline should be heavier than the generation pipeline. I encourage my team to use it as a plugin too. The "sorry way" is still a fine starting point — but once you see what a structured agent pipeline catches that manual review misses, it's hard to go back.