Those of us who have figured out the A.I tools have come around to the same conclusions; how to jump in, what to feed it, the setup needed. Read this article if you need some stronger guidelines and ways to really cook your agents, Effective harnesses for long-running agents by Anthropic. Those having a lower success rate are lacking the fundamentals of software building, system design. You have to have strong fundamentals. How else are you going to be able to understand the bottlenecks your A.I agent can get you into? Also soft skills in communication, you have to be able to communicate what you want. Formulate coherent thoughts and explain in detail where vagueness arises. Help the agent understand what you need done with clear directions in the smallest pieces possible. Don't give it vague and a large set of features or instructions to accomplish, learn to break down your problems into smallest form and feed it a quantity of good quality features.

I enjoy the freedom it brings, I don't need to be granular at the code level all the time. System design and architecture, nail those. Nail the UI/UX. Know your language or framework, ask the A.I. agent to explain sections of the code you don't understand. Think in a smaller scale and you can build both green and brown features for applications. Debugging can become a breeze. I can then multitask my deployment and look at options there, then quickly change course into database optimization. As fast as my agent can nail features together. I'm now multiple of me, I can work faster. I learned all of this by trial and error, even forget and still act hastily at times with my agents and get stuff wrong.

GROW, LEARN, HAVE FUN!