AI skills / continuous improvement

There's really only one thing you need to do to keep up with AI, and that's create and maintain documentation. Without documentation, AI will never be able to do meaningful work in your organization.

And with it, it can already handle most works done on a computer better than a human. I'm gonna talk about the two types of documentation you need and how you can start creating it now very quickly.

This is gonna be the abbreviated version of the much longer process we go through for all of our clients at our agency. The first kind of documentation you need is a company knowledge base.

If you want the AI to do any kind of meaningful work in your organization, it needs to understand everything about your organization. Collecting and structuring all that information takes a lot of time, but you don't have to do it all at once, and Claude can help.

Even if you're going to be using ChatGPT or Gemini for the actual work, you want to use Claude for putting together your documentation. And that's why this is such a good use of your time right now...

t to retrieve just the information you need for a given task. Then you're gonna give it your website URL, any social media handles, and any other documentation or data you have about your company.

Just dump everything in there. Claude is really good at handling large amounts of messy content.

One pro tip: instruct Claude to separate the documents into those that contain information that can be shared publicly and those that contain internal information. If you don't do this, you will probably find AI models are sharing internal information in public-facing content.

Once it's done creating the documents, ask Claude to create instructions for how to use the documents. This will just be a short description of each document and where it should be used, as well as which documents are for internal use only and which have information that can be shared publicly.

Now, these documents can be hosted locally or somewhere like Google Drive or Notion...

handling for you. The first skill you should build is one to update and maintain that knowledge base.

In that same chat conversation you had to create your knowledge base, just tell Claude: I now want you to make a skill for updating and maintaining this knowledge base. Whenever I give you new information about our company or industry, you should look through all of our current documents and determine if any of them need to be updated.

If the information doesn't fit in an existing document, you should create a new one. And that's all you have to do.

Claude knows everything it needs to create the skill. You can then just save that to your Claude account, and you can also download it to be used on other AI models.

And now, whenever you do any task, whether it's drafting an email, writing a press release, brainstorming strategy, you're gonna make sure you have a skill for that specific task. You don't just start the task by saying, we're gonna do this task, and at the end, you're gonna create a skill for doing that task in the teacher...

Knowing how to do something on a computer is becoming less valuable everyday, but knowing how to articulate how to do something on a computer is becoming more valuable. Google's release of Skills Inside of Chrome yesterday is a perfect example of this.

But before we talk about that, we need to talk about what skills are and why they're quickly becoming your most valuable business asset. In its most basic form, a skill is just text explaining how to do a specific task.

There's nothing technical about it. Anyone can write this.

If you've ever written a prompt before, you've already done 90% of the work. The idea of skills came from anthropic, the creators of Claude.

And still today, if you're looking to create some skills, which I highly recommend, you should be using Claude to do. However, all the major models, ChatGPT, Gemini, now also incorporate skills.

So if skills are just text with a set of instruction, what's so magical about them? Sure, I guess.

Now you don't have to write out those instructions every time you run the prompt...

e scope, the better. But even for a narrow task, the instructions, or what's called the skill dot m D file, might be 300 lines of text.

And then some skills will also contain what are called reference files. These are additional markdown files that might contain examples or guidelines.

Don't worry too much about the structure, though. Just tell Claude to create a skill, and it'll handle the structure.

But that's just the starting point. Then every time you use that skill, you're gonna provide feedback.

And have Claude incorporate that feedback into the skill. Do this 10 or 20 times, and you'll be surprised how capable it becomes.

But then the real magic comes from what are called orchestrator skills. Once you have a couple dozen narrowly defined skills, you can have orchestrator skills string them together into a more complex task.

At least right now. This seems to be the future of knowledge work, which is why we're seeing these skills get incorporated into every product, including skills for Chrome...

become easier with a skill like this. Or one you create yourself.

But the only reason this story is interesting to me personally is it confirms that this is the direction we're headed in. If you are waiting for a sign before investing a lot of your time into creating skills, this is it.

AI progress is rarely easy to predict, but it looks like skills, at least when they're built well, are gonna be valuable for at least the next two years.