One of the interesting things about the emerging AI-economy is where it is hitting. Obviously, two and a half years ago, ChatGPT opened up the space with a general-purpose app that offered users responses to any query that came to mind. It doesn’t quite fulfil that promise, but that was what was on offer.
At the other end of this is how AI is shaping the inputs to the digital economy: namely, code. There is a lot of discussion of ‘vibe coding’, which means making computer programs without knowing how to code. I suspect it is overblown. Like writing term papers, a single prompt or two will rarely do the job without some outside knowledge playing a role.
This causes me to wonder whether the future of the AI-economy is somewhere in between. This came to mind when I saw a new web app developed by my All Day TA co-founder and UT colleague, Kevin Bryan. The web app is an editor he coded up in order to help him edit various documents he writes, from papers to blog posts. You can actually use it here, so long as you supply a Google Gemini API key. Here is Kevin talking about it.
What’s the significance of this. Let’s start with how it uses AI. First of all, Kevin used AI to code the app itself. He laid out what he wanted and tweaked until he had what he needed. It took 4-5 hours I’m told.
Second, the app uses AI (hence the API key). It uses it to process documents and then to run various tasks.
So, on two dimensions, it wouldn’t exist without AI. AI was an input into creation. And AI was an input in being able to run the app itself. This dual nature of AI is what it makes it potentially interesting.
The significance of this is what it means for personalisation. There are many tools that can do all the things in this web app. However, this is the only tool that matches the workflow that Kevin likes to use. In other words, it is a user-generated app, and so it is extremely personalised. In that respect, it is not built to scale. It is built to be a tool for a market of one.
This is not new. Those who could code often built tools for themselves. What’s new is AI has allowed someone with far less skill (with all due respect to Kevin, “he’s an economist not a computer programmer!”) to do the same thing quickly.
And in terms of all the functionality, ChatGPT can do all of this. It just takes more copying and pasting and repetition. What’s different is how AI is accessed. It is accessed the way Kevin wanted to access it, rather than averaging out as happens with ChatGPT. It is basically the same philosophy behind the apps we have handed professors at All Day TA. We basically made a tool that allowed them, rather than OpenAI, to code their own assistant.
That said, it is possible the others have workflows similar to Kevin’s. To that end, you can just download the index file for his webapp and feed it to your favourite AI code editor and tweak it for your own preferences.
Pretty soon there will be thousands of similar examples. The question will be: will it be worthwhile searching for one that is close to what you are looking for or just building something yourself from scratch?