All Talk vs All Do
I wanted to say “Talk Sht vs Do Sht” but I’m trynna keep it family friendly. Working on LeeClaw and I want to ensure that the bot when finished can take meaningful actions (write code, do web searches, install what it needs to accomplish tasks, create websites).
The cool thing about writing your own bot is you get to define what it can do and how it accomplishes that. But - the downside - is you don’t have perfect knowledge of the code. Well, I have next to zero knowledge at this point. I’m acting like the product owner and tech lead.
I pursue ideas of MCP vs skills. I had a fuzzy idea what they are and that they aren’t the same thing. While walking, Gemini and I had a chat about them, giving me clarity. Basically you’ll create skills and direct features into your project to do local things and “must run quick” activities. MCP is when you want to call out to SaaS software and the like for scalable integration. Or so is my current understanding.
So step one is to do a code review. I discuss all this with ChatGPT and he writes a very detailed prompt to determine “where we are” in the “do sh*t” capability, and to define a maturity plan to move us forward. I give that to Codex Web as a PR (it is attached to my repo) - and I go walk the dog.
I come back and I have a detailed analysis with a sprint plan. I give the Sprint Plan to codex to code. Then I show the output to ChatGPT and ask “have we implemented all that the review and maturity prompt talked about”. Nope, so Sprint 23 plan is created.
I have Codex review the sprint plan and he reformats it to meet all our other sprint plans and then happily gets to coding - while I write up this post.
If this sounds a LOT like what I do in my role as tech lead, enterprise solutions architect in my day job — only with with me running ai coding agents instead of directing people - you are correct. I think that’s why this doesn’t feel like cheating or short cuts to me. I’m still doing all the thinking and problem solving and architecting that I always do. I’m just not writing the code.
Learning is a blast. The good ideas I’m incorporating into this project are coming from the pain points of the last few. I’m developing projects, the skill of using AI to develop projects, and a methodology to effectively use AI to build projects.
Come on in - the water’s warm.