If you’ve been on LinkedIn recently, then you may have seen the chatter about teaching your artificial intelligence to have various skills. I saw one post by a guy who claimed to have created several skills, each representing a tech billionaire.
At first, I thought “I am behind the 8-ball. What is this new thing?”. Obviously I know what the word “skill” is and how people use it, but I had not encountered its use in the context of AI having it. What does it mean for an AI to have a skill? I somewhat dreaded the the work of learning the new skill of teaching my AI skills.
Then I had lunch with a computer scientist and I learned that skills are nothing new.
Once upon a time in the 1990s, kids would sit at a desktop computer and just explore around. We’d change various settings, explore the directories, and try to figure out what the various file extensions meant. Quickly, I came to expect most installation packages to include at least one readme.txt file. The name was almost always readme. What was in it? Mostly boring stuff.
Readme.txt files still exist. They’re super basic files that appear in black and white and have minimal formatting options. They might include some version information or user agreements. But sometimes they also include really useful information about the functionality of an application. A readme file basically introduces a file or folder to the user. Sometimes the details are arcane and useless. Sometimes, they were indispensable for getting the most out of an application. Basically, if you read the readme file, you might easily learn some useful skills that were there thanks to some programmer’s foresight. That guy or gal wanted you to use the application to its fullest extent.
Now, if you view some code package or repo on GitHub, then you might notice readme files that have the *.md (markdown) extension rather than *.txt (text). Markdown files allow for text formatting that includes headings, lists, code blocks, images, etc. They tend to be relatively basic and may even include tutorials. Here’s an example from an investment portfolio program that I created.
If you’re using any brand of AI, then it can probably do a great variety of things with varying degrees of competence. If you want it to perform a specific detailed task, then you’d need to provide a specific and detailed prompt. Of course, that can be time-consuming. Even then, AI may run out of context if it’s a big project. Therefore, prompt engineering became a thing. People learned that a good prompt can make all of the difference in your project.
What does all of this have to do with AI skills?
Like I said, skills are nothing new. When a human reads a readme file, they learn something useful that is specific to the current application or task. AI does the same thing, albeit with a shorter memory. Rather than building a method or a practice from the ground up each time that you want AI to do something, you can instead create a readme file. If you’ve integrated AI into your code interpreter, then you’ve probably noticed when it updates its “memory.md” or “AGENTS.md” file. That’s the markdown file that it uses to store key pieces of information about you or your project. It helps the AI avoid redundant processing and dead ends. Instead, it can refer to the markdown file and have a sense for how it should behave.
A skill can be just plain text written conversationally, it can be a list of rules, mathematical expressions, or even the foundational code that you want your AI to readily modify and apply. Essentially, saying ‘skill’ is the same as saying ‘pre-prompting’ with various degrees of specificity. Rather than writing a prompt each time, you can recycle a set of prompts that you’ve stored in a file. That’s all that a skill is.
So, when you see people proudly say that they created one skill or another, 1) they are probably rightfully proud of the time saving that they have designed, 2) they are just creating a glorified readme file, 3) they are not being specific about the contents of the file. Anyone can create a markdown file and they can include a diversity of contents. There is no reason to be obscure, except to make it seem more magical. Depending on the author, looking under the hood can reveal a lot of careful pre-thinking, or a list of flashy but unimpressive bullet points.
Economist Writing Every Day 