Hey Reader,

I've been using Claude Skills for almost half a year now, and I'm happy to say that they have fundamentally changed how I work with AI, and I think they are here to stay no matter what kind of new AI model comes out.

They are:

1. Portable
2. Customizable
3. Easy to share

I just hopped off a conversation with one of our members, Paul. He has a real estate business in the UK, and he is not technical at all. He's been struggling setting up my Make.com template for content repurposing, and booked a free 1:1 call (one of our member perks) to get help with that.

And instead of setting up this whole system with him inside Make.com, which would be harder to maintain and harder to modify, I helped him set up the skill version instead.

He set it up in two minutes, and he had his first five LinkedIn posts written with AI using my skill in another two minutes (one minute of that was finding some source content, and another minute was waiting for the AI's response).

He asked me "are these Skills like GPTs?"

I said: "Yeah, but you can use multiple skills in the same chat thread, and your Skills will also work with Claude Cowork and Claude Code.

So for example, you can combine the Content Repuprosing skill with a Writing Style skill so it sounds more like you."

And to be honest with you dear reader, the effort you put into making good Skills for yourself is well worth it.

Skills are here to stay.

Even if we fast forward into the future where we have superintelligent AI helping businesses get things done, a Skill will still have value.

Because a Skill personalizes HOW things are done in your business.

(actually there are 2 kinds of skills, an SOP is one kind.)

I don't get why people are still sleeping on this. I think 90-95% of my AI usage is a combination of LLMs and Skills.

Heck, I even taught my mom how to use Skills, and she now has her own writing coach skill too, that helps her in writing the way SHE wants to be helped, and another one that handles her company car's trip tracking which she hates to do but has to due to Hungarian regulations.

If you think about it, a business can be simplified into being a combination of SOPs.

Good SOPs = Good delegation = Good business

Bad SOPs = Bad delegation = "Hard-to-grow" business

Skills are great, because you can easily create them (with the skill-creator skill), change them when your process changes, and they work with any LLM that supports Skills (for now Claude does this best, but Skills are an open standard, and if an AI company doesn't want to be left behind, they will implement this soon).

Where things fall short

Right now, I see two main roadblocks in how well people can use AI:

1. They suck at engineering the context.
2. The AI can't do the thing they ask.

Let me explain these briefly.

Context engineering is about making sure that the AI sees and uses the right information at the right time.

For example, that it knows enough about you and your business to have overall context, but it's not overloaded with a bunch of irrelevant "project memories" or stuff like that.

Agent harnesses like Claude Cowork and Code require more of this context engineering, so if you work with those (and I recommend you at least try Cowork), I recommend that you have one main folder called a "Claude Cowork & Code folder", which you use for all kinds of AI related work.

In that folder, on the top level, you should have:

• A CLAUDE.md file which provides overall context about you and your business. Claude reads this EVERY time you start it in this folder.
• a Projects folder, where your projects are categorized into subfolders. (for good measure, have an "_archive" folder with older projects nested in there)
  • Each project folder should also have it's own CLAUDE.md file with project related context. So when you start Claude in the top folder and ask it to go into "projects/example-project", it reads BOTH the project specific CLAUDE.md and your main CLAUDE.md one folder up.
• a Skills folder (this should live inside your .claude secret folder created automatically) for your skills
• a Documents folder for storing documents not related to any project. (this is basically a Misc folder for all kinds of crap)

If you have this structure, you are off to a good start. Over time, your structure will grow, but tthis is the bare minimum I recommend (I now have other folders in my root folder, like "extensions", "screenshots", "plugins", "scrips", "yt-transcripts", etc.)

It makes context engineering much easier, because:

• you can tell Claude Cowork (or Code) which project you're working on and which skill to use
• projects will not cross-contaminate, because Claude only reads the files in another project when you tell it to do so.

If you have such a "working folder", you don't need to worry about any other website feature, like Claude Projects or Custom GPTs, or whatever.

And when a new model comes out (like Opus 4.7 just did last week), you can utilize it's full power instantly because everything is set up.

Skills also help with the context engineering, because you can reference certain filepaths in your Skills, like "you can find screenshots in workspace/screenshots".

Actually, it's not even engineering, it's more like management.

And the other roadblock on why people can't use AI to their full potential is that the AI can't do the thing, or not that well.

The list would be too long for me to collect everything, but generally speaking, where I see most issues with AI (or where people get disappointed about AI) are:

• Working with images: vision is still a hit and miss, some AI tools like Claude can't edit images at all, whereas ChatGPT or Gemini can, and therefor I think people assume all AI can.
• Uncommon fileformats or files falling apart: LLMs can still only handle plain text. So anything that's extra, the AI needs to convert it to plain text, which is still a hit and miss.
• AI not knowing about it's own limitations, so it says it can do something, but when you try, it can't, so you think you did something wrong.

On the other hand, AI is insanely good at a lot of things now that people don't expect:

• Handling complex projects: AI can spin up subagents and delegate tasks to manage context, it can compact the conversation to keep going, it also has longer context windows (Opus 4.6 & 4.7 has 1M which is more than enough).
• Calling tools and getting things done: I keep getting surprised by how well Opus can call an MCP server or write a python script for itself to call an API and get shit done
• Making and organizing files: I don't think most people actually understand that every file they have ever worked with is just computer code that an AI can read and write. Even a 3D printer .stl file is just code. A video is just code.

And if an AI sucks at reading or writing a highly specific file format you regularly work with:

Just teach it how to do it with a Skill.

(this is the other type of Skill, which extends the capabilities of the AI in a field it hasn't been trained in.)

Dave Talas

P.S: I talk a lot about Skills and also shared quite a few of my own in my AI implementation community for entrepreneurs, The Paladin Guild. It's currently closed for enrollments but join the waitlist, and I'll let you know when it opens again!

P.P.S: Also it's crazy how I asked claude to "come up with a way so that I can quickly open and close enrollment for the community" and it created such an elegant solution that it takes me 2 seconds to open or close the cart. I just have to modify a variable from TRUE to FALSE and redeploy, and it loads either the checkout link or the waitlist form.

In the meantime, check out my YouTube channel for AI related videos. Is there a specific topic you'd like me to make a video about? Just reply via email and let me know!