In this week's note, I'm taking a break from my Donella Meadows deep dive series. I want to share something I learned recently.

Over the weekend, I tried to move my LLM Wiki workflow from OpenCode (with DeepSeek) to Codex. The workflow was already working. I wasn't trying to redesign it.

What I wanted was simple. AI reads and interprets the information, then creates wiki entries. I review those entries and decide what to keep, change, or reject.

The migration started smoothly. But as I worked through the process with ChatGPT, things started getting more complicated.

Instead of preserving the existing workflow, ChatGPT started proposing a redesign. Additional logging systems, fingerprints, approval stages, and various safeguards.

Some of the suggestions made sense individually.

But did I need them?

This reminded me of an old problem.

I've built many personal systems. One trap I kept falling into was continuously "improving" them. Adding feature after feature.

Now, AI can build many of the new features I could think of much more easily.

But am I building a system to improve my workflow and help me learn better? Or am I just building for the sake of building?

In the past, I've had apps and workflows that became so complicated that I eventually stopped using them.

There was too much friction just to complete a simple task.

So I pushed back against the additional processes. I returned to what really mattered for my LLM Wiki workflow.

AI reads, interprets, and writes. I review and decide.

AI has made building software much easier. Adding another feature or process takes far less effort than it used to.

But I still have to live with what I've built.

And that's something I need to remember before adding the next feature.

Until next time, keep learning.
Gav