There are two types of computer users: those who use devices as-is out of the box and those who hack about to get the most out of them. The first group might install some apps on their devices, but the second group goes further by learning to automate things via scripts. LLMs have opened a golden age for the second group.

I’m in the second group. I’ve long appreciated the possibilities of automation frameworks. The Mac (my primary platform) has several: Shortcuts, Automator, AppleScript, and the Unix shell environment. By using one (or more likely, several) of these frameworks, I can get my computer to help with a bunch of menial tasks.

What kinds of tasks? Mostly repetitive things (e.g., renaming hundreds of files, monitoring a directory to archive files in a particular subdirectory structure) or workflows that string together several apps (e.g., using one app to resize and reformat image files and another to save them to a directory in Github.)

Much of what we do with computers are repeatable patterns. By letting the computer go at it, these frameworks relieve us of tedium and make us more efficient.

The downside is that they’re not easy to learn. Writing Unix scripts or Shortcut workflows isn’t something one does every day. As a result, most people don’t develop the expertise necessary to be super productive with the frameworks. We stumble along by trial and error, which takes a lot of time.

In the before times, we cobbled together the knowledge to do this by Googling for directions. There are general programming websites such as StackOverflow and specialized websites that focus on particular frameworks. They all offer examples that serve as a starting point.

But most examples are just that – examples. They’re not the solution to the exact problem you’re trying to solve. At best, they provide ideas you can map on to your problem.

LLMs change this process: you can tell ChatGPT what you’re trying to do and ask it to give you a script or Shortcut workflows and it’ll write a very competent first draft for you. It won’t implement it for you – after all, it can’t control your computer (yet) – but it’ll give good directions on what to do.

No more need to spend dozens of hours experimenting to get things right. With the LLM, you get 80% of the way there in minutes. From there, you can tweak things to flesh out the remaining 20%. The new workflow opens up lots of possibilities.

That said, there’s a downside. I’m comfortable having LLMs draft automations because I’ve used these frameworks for years. I understand how they work. But I can imagine a complete neophyte getting in trouble with this process. With great power comes great responsibility, etc.

Still, for prosumers who understand automation, working with LLMs changes what we can do with our computers. I’ve already improved a several workflows using this approach and am working on many more. Give it a shot if you’re so inclined.