I Directed an App!

A few weeks ago, my colleague Michiel, our CMO, showed me how to use Claude Code to work on the company website.

It blew my mind what Claude Code could do, especially when you put it in its highest-effort mode. It would find and take into account things that were relatively hidden in the repository. It would “think” through implications of changes that I hadn’t considered. What I thought was a simple copy change on one page, for example, triggered a response along the lines of: “Done, but have you considered how this affects similar copy on other pages?” I hadn’t seen this level of “thinking” from an AI before.

That reignited my interest in AI.

Only a year earlier, I had worked on Strom, a Python script for automatically managing heaters, for which my friends and I also used AI to write most of the code. That, however, was a pretty frustrating experience. What a difference a year makes. Back then, there were no effort levels, to my recollection, the models were worse, and on top of that we were limited to the free plans.

We got Strom to work in the end, which was very satisfying, but the road there required a lot of patience and trying a million things until we found the one path that worked.

In the past few weeks, I’ve been using the new models at their higher effort levels, and it’s a completely different story. It’s really hard to put into words how good they have become.

One day, after my Chinese class, I thought I’d scan my class notes and see what ChatGPT’s current best model, 5.6 Sol, could do with them. The results, again, blew my mind. It created a great flashcard web app. The app, creatively named Yuwenke, which means Chinese class, is nice, fast, responsive, and personalized to my class.

The first version took around an hour to create and was already very usable. Then I connected Yuwenke to Firebase so that users could save their progress by signing in with a Google account.

As a user, this is so much better for me than Duolingo. There are no ads and no giant app taking up space on my phone. Yuwenke is 100% relevant to my in-person class. And I can change it whenever I think of a new feature I’d like to have. Yuwenke was free for me to create and is free to run, except for the €20 ChatGPT subscription I used to create it.

It’s hard for me to consider all the implications of this new reality. For one, it means that when it comes to the simpler apps I use, if I don’t like them, I can just create my own version.

By the way, what is the correct verb here? I sure as hell didn’t program this app. Did I create it? More or less. John Gruber advocates calling this act directing and compares it to the work of film directors (The Talk Show episode 7952 at around 1 hour 20 minutes). John argues that film directors are not necessarily personally involved in every aspect of making a movie. What they do is hold responsibility, give direction, care, and answer questions. That’s exactly what this new class of AI-assisted creators does, and what I did with Yuwenke.

If it’s hard to think through the implications for my personal and professional life, I find it even harder to begin thinking through the implications at the scale of entire economies. What happens when more and more people become aware of these new possibilities?

Interviewed by Yascha Mounk, Arvind Narayanan says (via Arnold Kling):

any task in the economy that you can specify precisely enough for it to even be legible as a task is going to be done by AI. …What jobs will mean in the future is what economists sometimes call interstitial tasks: the things that lack a precise specification, where maybe part of the task is even to figure out what should be done next, things that maybe have never been done before. Those are the kinds of things we think will be in the domain of humans. The reason I think we will be very unlikely to leave those things to AI is because of unknown unknowns. You can’t be sure that when you task AI with doing something so different from anything that’s been done before, it’s going to do a halfway reasonable job.

The importance of things that lack a precise specification is why the position of director is already considered essential in some industries. Everyone could, in theory, make a movie with their phone nowadays. Phones generally have great cameras and enough power to edit videos. Everyone could, in theory, follow a set of instructions on how to create a movie and make a movie. But, of course, there is way more to making a great movie and much of it lacks a precisise specification.

My initial gut feeling is that directors, the people with good judgement who take care of things that lack precicise specification, will become more and more valuable. When everyone can create an app in 30 minutes, making a great app in the right way becomes the differentiating factor. The same goes for every SaaS business. Competition is heating up, everyone with a laptop can spin up a SaaS business with a functional platform in an afternoon. Having a business-Scorsese directing a SaaS is going to become increasingly important as techical execution becomes less of a barrier to entry.