LLMs and Institutional Knowledge

Baetjer’s Software as Capital (sadly, long out of print) introduced me to the idea of institutional knowledge. As the author explained, nobody at General Motors knows how to build a car; rather, it is the organization as a whole that knows, and while the value of that “distributed knowing” is hard to quantify, it is very real.

After an hour and a half with Jon Udell yesterday using ChatGPT and Claude to solve a CSS problem, I now wonder if LLMs are a way to package institutional knowledge for redistribution in the way that Docker is a way to package a developer’s desktop environment for others to use. Stack Overflow was a big step in that direction, but only worked if someone had asked your question before (or a question close to yours in your current knowledge space). While some of the LLMs’ suggestions yesterday were wrong, others were close enough to right to save me several hours of frustration. I’m still very afraid of what will happen if we don’t enact a regulatory framework with real teeth (i.e., that can and do inflict significant financial pain on careless or uncaring companies), but I gotta say, brew install helpful is pretty appealing…