In my last post, I wrote about taking a naming expert's process and turning it into an AI skill. I was happy with the name it came up with. But I also spent quite a bit of that post thinking about what this means for people who share their expertise.
Here I want to get into the practical side. What are these skills, and how can you use them in your business?
What is a skill?
A skill is a set of instructions your AI tool can reuse for a particular job. It can also include examples, templates and code that help it do that job.
Think of your AI as a very smart employee who knows nothing about your business. You still have to teach them how you operate, what role you'd like them to play and how you want the job done. Being smart doesn't mean they know your way of doing things.
Now imagine this employee may also have amnesia. You ask them for something you've already explained, and you find yourself explaining the whole job again. What you told an AI in one conversation may not be there in the next.
This is where skills fit in nicely. You write down how to get the job done so your AI has instructions it can refer to when it needs them. You may still need to tell it which job you're working on and give it the relevant information, but you don't have to explain the whole process from scratch each time.
Okay, so what should you do with that? There are two uses I want to talk about.
Work you keep doing over and over
Think about things like monthly reports. You log into multiple dashboards, consolidate the numbers, create a chart, an Excel spreadsheet, a presentation perhaps, depending on who you're presenting that report to. Then you share it.
That's a set of steps. If you can explain those steps clearly, and a computer can follow them, you have something you could turn into a skill.
You would need to tell it where the numbers come from, which dates to use, how you calculate things and what the finished report should look like. The AI also needs access to the right tools or files. Writing instructions doesn't give it access to your dashboards.
And you still have to check the result. Following the same instructions doesn't mean it will get everything right every time. But you have a process you can test and improve, instead of starting the explanation again next month.
Getting a little bit of domain expertise
The other use is what I wrote about in the last post. I listened to David Placek talk about how his team comes up with names, then used that approach to create a naming skill.
Now, when I read a book or listen to an expert, that knowledge is in my head. For the rest of my team to use it, I have to explain it to them. Everybody will have different takeaways, and we won't necessarily apply it the same way.
What skills allow you to do is turn parts of that knowledge into repeatable instructions. Something you can maintain and update, and that other people in your business can use with their AI tools.
It won't be the same as working with the expert directly. But you can take what you've learned, adapt it for your use case and give your AI a more useful process to follow.
Over time, you have a folder of these skills that you keep adding to or updating. The things you learn have somewhere to go beyond your own head.
A word about using other people's skills
Before you start downloading skills, there is something important to understand. These are instructions, sometimes with code attached. They can tell an AI to do things you didn't intend, including accessing files or sending information somewhere else, depending on what you've allowed the AI to access.
So don't take a skill from someone else and run it blindly. That includes the naming skill I shared in my last post.
You should look at other people's skills and learn from them. But understand what you're using and adapt it for your business. Asking an AI to explain or review a skill can help; it doesn't prove that the skill is safe. If you can't assess what it does, get someone who can before giving it access to your business data. Anthropic has guidance on this here.
Where to start
Pick one task you keep repeating. Write down how you do it, including the parts you normally leave out because they seem obvious to you.
Ask your AI to help turn those steps into reusable instructions. Give it an example of what a good result looks like. Try it with sample data first, check what comes back, and update the instructions where it gets things wrong.
Start with something small enough that you can tell whether it worked. Then you have something to build on.