The Pi-agent harness can handle Codex subscription plans, so using that will make things go much faster. It's a minimal harness designed for extensibility, but because it doesn't add unnecessary system prompts, it unleashes the full potential of GPT and improves problem-solving capabilities.
In my case, I had a problem with OpenCode that I was using a Codex subscription for, and it took me 6 hours to solve, but I couldn't solve it all. However, I solved it in 20 minutes using pi-agent.
https://pi.dev/
