Inception Mirror Land
I’m Bruce Lee - and I have entered the hall of mirrors, chasing after Lord of Productivity…whom I must overcome and master! Ok, a little dramatic. I am working on several layers of the stack across a few projects simultaneously. What do I mean?
I am working on a project as a way of learning a tool. I’m learning a tool and how to use it productively. I am developing and refining the methodology for the tool. On more than one project, simultaneously.
At the risk of indulging those who accuse me of having AI write FOR me…I can’t help but share this line from Gemini that it came up with while discussing this post (that I am writing myself):
Gemini: It feels like we’ve moved past the era of “I’m using a hammer to build a house” and into "I’m using a sentient hammer to design a better hammer while it builds three houses at once.”
The projects:
1. Book writing/AI tutorial. For my daughter, a complete guide from beginner to mastery for how to write a book, and how to use AI. The goal is for her to learn AI tools while learning how to write a book. And I’m using this project to develop an autonomous, continuously improving process. Inception, mirror land #1
2. OpenClaw - yes, I’m back at it trying their latest version. I’m using Codex to supervise the installation, security hardening and best practice setup to get an autonomous agent that can accomplish tasks on my behalf. I’ve fed Codex cli several best practice tutorials, and it has synthesized them and made a plan. So I have an AI, building and following a plan to install and configure an AI bot to autonomously accomplish tasks. Inception, mirror land #2
3. Integrating OpenAI’s Symphony project an open source app they have created to coordinate a swarm of AI agents working on tasks.
Gemini writes: Symphony isn't just a "swarm" in the sense of a bunch of chatbots talking. It’s built around "Harness Engineering." * The Workflow: It watches a task board (like Linear), spawns an isolated environment when a ticket is ready, and assigns an agent to solve it.
- The Inception Layer: The agent doesn't just write code; it runs CI tests, handles its own PR feedback, and—this is very "meta"—generates a walkthrough video of its work for you to review.
- Your Twist: You are using an AI (Codex) to set up and supervise an AI framework (Symphony) that will then spawn its own AI agents.
4. In the process of all this I’m learning about the use of Skills - think of these as a combination of functions and include files in c. They are stored prompts that come into play when needed but aren’t taking up llm context when now. I’m also working on creating my own orchestrator to overcome the weaknesses of non-deterministic prompts. You’ll experience this as “I told the llm to do a, then b, then c.” But it decided to skip b for some unknown reason. An orchestrator layer can ensure all the steps are run in the proper order with valid artifacts.
I’m bound and determined to get my $200 out of OpenAI that I signed up for a week or so ago.
Yes, it’s hard to keep all of these projects straight.
Yes - it’s a BLAST! I may not be on the bleeding edge, but these areas I’m working on are what a LOT of people are. We are hitting the weaknesses of LLMs and engineering solutions to those weaknesses.
AI, agents, bots - these are not the end of human ingenuity and thinking. They are not the end of coding. They are the engine increasing the power of creativity.
Oh crap, forgot to add I’m also recreating Load Runner - the version that ran on the original Mac. I’ve enjoyed playing with my Space Invader’s Deluxe game, and now I’m attempting a new one.
What I’m really doing right now is learning how to build the machinery that lets AI build things reliably.
