READY

So what entails competition? Are we building something?? Sorry, I'm so overwhelmed with everything lol

I’ve spent time experimenting with a wide range of AI tools and agent harnesses, including Claude Desktop/Co-work/Code, ChatGPT/Codex/Codex CLI, Antigravity IDE/2.0/ CLI, Cursor IDE/CLI, Hermes, Pi Agent, Aider with locally hosted DeepSeek V4 Flash, as well as several lesser-known projects. I’ve also built my own agent harnesses to better understand the space from the ground up. After all of that, one conclusion stands out: every harness has its own strengths and weaknesses. Some are purpose built for particular use cases and might different in that aspect from others. There is no universal winner. In my experience, the quality of the framework matters far more than the tool itself. Context engineering is ultimately what determines the outcome. The teams and individuals who can effectively structure context, orchestrate workflows, and manage feedback loops will consistently outperform those who simply chase the latest model or harness. Personally, I prefer using multiple models/agents in specialized roles, collaborating within the same workspace and reviewing each other’s outputs. I believe strongly in checks and balances, and I’ve found that the same principle applies surprisingly well to agentic systems. Having models challenge assumptions, validate outputs, and provide alternative perspectives often produces much better results than relying on a single agent. I started my journey heavily invested in Claude, but over time I’ve become increasingly reliant on local models to automate business workflows and maintain greater control over my stack. While Claude remains an excellent product, some of Anthropic’s decisions and inconsistencies have made me less enthusiastic about building around their ecosystem. That’s simply my personal view—no hate intended. Claude Code is arguably the easiest agent harness to get started with today. However, from a value perspective, I currently find Cursor and Codex to offer a stronger return on investment among mainstream agents. Of course, this space changes almost daily, so ask me again in a month and I may have a completely different answer. If I were you, I would start with Cursor, Claude is too overrated in my opinion. With Cursor, you get to at least try out different models and use it for their strengths.

Hey guys, please review this and let me know your thoughts.

@Greg Faysash That’s awesome! I’m glad we all are rediscovering software engineering through AI with our own frameworks. My skills and conventions reside in the vault as well. I built a common knowledge base for all my agents. The skill that I was talking about is for the agents to remember my collab structure. This is to reduce the probabilistic and erratic nature of AI by reminding my protocol every time they see it.

I actually used to structure my workspace even before I encountered your method but your method helped me with organization. I took that learning and enhanced it and started engineering loops with disparate agents even before loop engineering became a thing on social media. I think to your point, we are rediscovering software engineering but with AI.