Listening to the Paul Roetzer and Mike Kaput podcast, The Artificial Intelligence Show, from three weeks back (it’s been a busy few weeks), on the ride to work this morning, I realized that Mike’s comment about how regular people are using tools like Claude Code and ChatGPT Codex applied to me. Regular people, in this case, isn’t using these command-line tools (I’m on MXLinux so that’s the only way I use Code and Codex these days, as opposed to the GUI available on Windows and Mac, although I have occasionally used those) for coding, but for organizing and preparing information.
My Latest Projects
One of my latest projects, of course, has been using ChatGPT Codex for cleaning THOUSANDS (6K-7K) of WordPress.com blog export files. Everything from images, to blog entries, cleaning up and throwing out old text, addendums to blog entries that made sense back when I was working as a K-12 technology director in a large, urban, political school district. Now, a lot of that cruft needs to be cleaned out. I also had images pointing to WordPress.com, and in some cases, BlogSpot (I don’t think I moved those images over…too many useless illustration images for over 7000 blog entries at Around the Corner).
I would NEVER have been able to clean out all that content myself. There’s just no way. I probably would have nuked it, and started a new blog. But Codex was able to go through all that content and make sense of it, then make suggestions. Along the way, it engaged in parallel problem-solving.
Parallel Problem-Solving Through Decomposition and Synthesis
The idea of parallel problem-solving is one that I got from Dr. Judi Harris back in the 1990s. It’s the idea that you give students a problem, and then have different groups working to solve it. The solutions they come up with are different (maybe), and you get to showcase how they arrived at those solutions. What Gen AI does is a little different than that idea, but this is my brain making a connection. I’m sure there’s a term for this, maybe decomposition, where a large task is broken into smaller ones then assembly occurs, and then each individual item is synthesized into one.
A Diagram
Today, with Gen AI, I see tools like Codex and Code (which I’ve used to do some MASSIVE projects) gathering, organizing, curating, blending content after I’ve set up the parameters of what I want. Let me see if I can show this in a diagram:
graph TD
A[Introduce Goal or Task for AI] --> B{AI Analyzes goal/task, then assigns work}
B -->|Agent 1| C[Works on SubTask #1 of the Goal] --> G[Solution integration]
B -->|Agent 2| D[Works on SubTask #2 of the Goal] --> G[Solution integration]
B -->|Agent 3| E[Works on SubTask #3 of the Goal] --> G[Solution integration]
B -->|Agent 4| F[Works on SubTask #4 of the Goal] --> G[Solution integration and Presentation]
G --> H{Human reviews solution and suggests modifications with screenshots, copied and pasted text to the AI to make adjustments}
H --> I{Gen AI produces the "final" solution and human accepts it}
An Alternate Diagram
The diagram above is my version…below is ChatGPT’s take on what my diagram should look like:
flowchart TD
A["Human defines the goal, requirements, and constraints"]
B{"Gen AI analyzes the goal and decomposes the work"}
C["Agent 1 completes Subtask 1"]
D["Agent 2 completes Subtask 2"]
E["Agent 3 completes Subtask 3"]
F["Agent 4 completes Subtask 4"]
G["Integrate and synthesize the results"]
H["Generate and present a proposed solution"]
I{"Human reviews the solution"}
J["Human provides corrections, examples, screenshots, or additional context"]
K["Gen AI revises the solution"]
L{"Human accepts the solution?"}
M["Final solution"]
A --> B
B -->|Assigns work| C
B -->|Assigns work| D
B -->|Assigns work| E
B -->|Assigns work| F
C --> G
D --> G
E --> G
F --> G
G --> H
H --> I
I -->|Changes needed| J
J --> K
K --> L
L -->|No| J
L -->|Yes| M
I -->|No changes needed| M