I wanted to productize some of my work with AI. There is good news and bad news…
AI can write code well because somebody has written something similar before. AI can write words well because there are so many well-written materials out there. AI can draw images relatively well for the same reason. AI can predict, with some probability, what someone might decide given certain circumstances.
I found the edge of that recently, working with a small security company. I was helping install an operating model built around what AI can already do today. And it’s not just about dramatically speeding up software development, but speeding up work in general.
During the process, I kept answering what seemed like the same questions over and over. The questions felt repetitive. The answers felt half-automated.
So naturally, I had a “fresh” idea: Productise it!
At the end of the project, I asked the LLM I was working with to summarise what we had done so I could think about productising it later. It produced a nice Markdown file. I saved it and, of course, never read it.
The next day, determined not to put it off, I gave the file to another LLM and asked: What could be productised there?
And there was something. Just not what I hoped for.
Those “same” questions turned out to be key differentiation questions.
Those “half-automated” answers turned out to depend on judgment. In other words, knowing what matters, why it matters, what is different this time, and what to do about it.
The repetitive part wasn’t actually repetitive. It just felt that way. That was where much of my expertise actually was. Apparently, the valuable part.
AI can do a lot with what has been done before. But it cannot reproduce your judgment precisely without somehow capturing everything your judgment depends on.
So there is good news and bad news at the moment.
The bad news is that you cannot automate your judgment precisely.
And the good news is that you cannot automate your judgment precisely.