20 minutes is all it takes to learn AI

Timestamps

• 00:00 - intro
• 00:46 - ChatGPT or Claude?
• 02:07 - Setting up your model using memory
• 04:04 - Skills
• 04:52 - How to create your own skills
• 07:28 - An example using skills
• 12:06 - Other tools
• 12:38 - Claude Cowork
• 17:14 - Getting started with Claude Code
• 20:12 - What's next?

There are a lot of different players when it comes to AI. ChatGPT is a big one, Claude is another, there's Grok and a bunch of others you're going to hear about. Which one you choose probably doesn't make all that much of a difference.

I use Claude for the majority of my work, so that's what everything here is built around. Claude currently does not have a free version though, so if you want to follow along without signing up or paying for anything, try ChatGPT instead. You can also download local models if you want to go that route, but most people who want to start using AI don't start there.

Whatever you pick, download the desktop app instead of living in the browser. Search for it, download, run it. It looks pretty much the same as the website, but running it as an app on your computer is an overall better experience. https://claude.ai/download

Projects

Most people start in the default chat window. I'm not going to recommend that.

Create a project instead. A project is where you define what you're working on, and every chat you start inside it gets organized under it. You can pin the project and it shows up in the pinned section on the side.

The magic is further down the page. Scroll and you'll find instructions, memory, and context. A default chat window knows nothing about you or what you're trying to accomplish, so you restart from zero every single time you open it. In a project, you write the instructions once: what you're doing, what you want to accomplish, how you want the chat to act.

Context is any document it might need to reference. Say you're working on something for work and it has to stick to a product spec or some terms of service, you drop that in the project context and it's available in every conversation there. Spreadsheets, documents, photos, same deal. If it's a one-off file you don't need everywhere, just attach it to the single chat instead.

The project doesn't have to be work either. Budgeting, some weight loss goal, organization, whatever you're actually trying to get done.

Connectors live behind the plus button in the chat. I have one hooked up to Course Catalyst, which is the platform I built, and that's how you get the AI interacting with other services.

Skills

A skill is how you organize a task you're going to do repetitively. It's basically just a way to copy and paste some instructions whenever you need them.

There's a browse screen with a bunch of available skills, mostly from Anthropic, who make Claude. A web artifacts builder if you want it generating HTML pages, a morning brief, that kind of thing. You can find skills online and install those too.

What I like to do is create my own, and the easiest way is to just ask for one. There's a skill creator skill, but you don't really need it.

A dumb first skill

I told it:

Whenever I say hello and invoke the skill, say the date and hello Caleb.

That's it as far as setup goes. It writes the skill as a markdown file, which is just a way to structure text, describing what it's supposed to do and the format it's supposed to be in. Save it, then type /hello in a chat and it loads the skill and gives back the date and hello Caleb.

Totally useless, and that's the point. The example is there to show you how it wires up, then you build on it.

One that's actually useful

Second one, still simple. I wanted to hand the AI a pile of information and have it split that out in to a to-do list, notes to keep track of, and blockers I'm waiting on.

I'm organizationally incompetent, effectively, so having AI do that part for me is a big deal. The prompt I gave it was to repeat back a brain dump from me, split in to to-dos, things to remember, and blockers, and to update the list on later messages telling me what changed.

Then I dictated a mess at it: I need to clean my house but I can't because I'm waiting on cleaning supplies, I spent $712 on cleaning fees last month and I want to track whether that improves, and I need to pick up a plunger and some pizza. It came back with the to-dos, the note about the $712, and the blocker on the cleaning supplies.

💡 Tip[!tip]
Sometimes Claude picks up on the skill and just runs it, but I usually prefer to be explicit and invoke it by name, so I type /organize myself.

Now, the thing you should know is that this does not persist between conversations. Unless Claude decides to write something to its memory, it lives in that one chat. So the next logical step is storing that information somewhere, a file, a database, Notion, something like that, and you can get AI to build that part too. If you're not sure what to do next, ask it, "how could we persist this between sessions?" You don't have to know the answer ahead of time.

Skills are one of the biggest unlocks with AI, and also one of the simplest things in Claude. Pretty much any capability you want, you can describe inside a skill, and you can publish it and share it with other people. So when you see someone online talking about how they got Claude to do X, there's a very good chance it's just a skill with instructions telling the LLM how to do that thing. Someone took the time to write down how to design a web page, or they had Claude help them write it down, and now it's a design skill.

Cowork

There's a middle ground between the basic chat and Claude Code, and it's worth knowing about even though I don't personally use it much.

Inside a project I asked it to create a file that can persist across sessions, then started a new chat in that project using cowork and said "read project state and make an example edit to it". The file is project-state.md, sitting in a folder, and the new chat read it and changed it. So you can work on something over time across different chat windows.

That's the actual benefit of Claude Code too, working on your own files so the stuff persists. Checklists, SOPs, configurations you update as you go. Instead of one-shotting a chat and hoping you get the right result, which is what a lot of people do with AI and why they don't get great results, you keep improving the same thing.

The next step from there is scheduling. The example it suggests is a daily briefing, and I think that's a strong starting point, because it isn't doing anything on your behalf, it's just reading and telling you what's going on. Emails you need to reply to right away, leads on social you forgot to reach out to, GitHub issues worth reviewing. You can run it manually or on a recurring basis.

I don't have a daily briefing set up right now. I have something in the works, but it's not where I want it to be yet, so that's probably future content.

Claude Code

This is the next step in the progression, and I already have a dedicated video on it that runs about an hour, so I'll point you there for the details: One Hour of Claude Code. This video here is the precursor to that one.

To get started, the installation command is in the docs at https://docs.claude.com/en/docs/claude-code/overview. I'm not actually sure whether it comes installed with the desktop GUI, it might, so you could just try it first.

I like to run it from a folder:

mkdir my-project
cd my-project
claude

If it's installed, it works. If not, run the install command from the docs.

Inside, you can toggle between modes, and eventually you'll graduate to auto mode so you're not approving every permission (I cover setting that up in the other video). Everything it generates goes in the folder you launched from.

You don't need a software development background for this, and you don't have to use it for coding. It can generate websites, apps, markdown files, CSV, Excel. It talks to other apps, APIs, and MCP servers, and works with pretty much every file type. You'll get a lot more done this way.

Quick example: my thumbnail. Not the most amazing thumbnail you've ever seen, but it wasn't "hey, generate me a thumbnail" either. The styling and the positioning of all that data are anchored in HTML, so I don't just have a PNG sitting there, I have the file that produced it and I can open it in VS Code and change any of it. That's way better than going to ChatGPT and asking for an image. I'm just using Claude Code to code, it just happens that what it's coding is an image.

There's also a remote control option, which puts the session in the code tab in the Claude app on your phone. That's what I do when I have something bigger, fire off a task that might take an hour or two and check in from my phone while I go to the store or go have a life outside of sitting at the computer. I say that, but I usually just sit at the computer.

What's next

Start using AI every day and figure out what it can do for you. Build out workflows for the things you already do, and go research what other people are doing with it. I follow a bunch of people on social media and I'm constantly getting videos of "here's how I used Claude Code to do this", and that gives me ideas to pull back in to my own workflows.

Learning the actual tools and the infrastructure goes a long ways, so you're not stuck in basic prompting. I'll have subagents working for multiple hours on different features. More on all of that in the hour-long Claude Code video.