Intro

Please introduce yourself:
• Your preferred name
• Your year (i.e. if you’re doing thesis)
• What media you’re interested in exploring
• What software/programming language you use the most

OBJECTIVE

Learn Claude Code :)

OBJEcTIVE

Understand how LLMs work, and how to get the best out of them for various tasks (research, writing, coding, and more).

OBJECTIVE

Learn to use the terminal! 
You might hate it at first, and you may never love it, but you will learn to accept it.

OBJECTIVE

Learn to extend Claude Code through MCPs, skills, and other tool usage.

OBJECTIVE

Learn how to use Claude Code to manage existing tools, extend them or build your own systems to support your art practice.

OBJECTIVE

Use the eval loop in Claude Code to build working software with confidence.

OBJECTIVE

Learn when to use agents in your own research and software. Then learn how to leverage agents in your own software.

SCHEDULE

Week 1 Install and Claude Basics
Week 2 Research and Writing with Claude
Week 3 Skills, MCPs, and extending open source projects
Week 4 Build your own software part 1
Week 5 Build your own software part 2
Week 6 Build your own software part 3
Week 7 Demo Day

CLAUDE CODE

In the past ~3 months, nearly the entire technology process that supports my artistic practice has changed. 
I think yours will too.

CLAUDE CODE

Regardless of your current technical skills, you soon will have the ability to customize your current software, or build completely custom software.

CLAUDE CODE

For many creative technology projects, software/creative development was the most costly/time intensive part of the project. The cost of that will now approach zero.

ClAUDE CODE

That means you can now create more projects, faster.

ClAUDE CODE

But: if your plan after ITP was to become a creative technologist for other artists, you now have a lot of competition :(

CLAUDE CODE

This means ideas are even more important. And your ability to connect ideas to technical processes is still a key skill.
But you might not be writing the code now. Rather you will plan the code and an LLM will produce it.

CLAUDE CODE

Once you start using Claude Code, you’ll begin to think about turning everything you do into custom software.

CLAUDE CODE

In fact, these slides are made with a Claude Code built application :)

INSTALL CLAUDE CODE

First, let’s get you your Claude subscription.
Thank Anthropic for two months of 5× Pro :)

INSTALL CLAUDe CODE

Before we install Claude Code, we need to get Terminal running.

Install Claude code

There are many Terminal apps, but I recommend:
• Built in Terminal
• Warp
• GhosTTY
• you can use an IDE too, I guess

install claude code

Claude Code installation instructions here

LLMS & coding Systems

“LLMs are just next token predictors”
If that’s true, why are they working so well now?

LLMS & CLAUDE CODE

I’d argue—bigger than any model change—the biggest improvement in LLMs has been the development of “the harness”.

LLMS & CLAUDE CODE

“The harness” is a way to describe everything around the LLM itself. 

LLMS & CLAUDE CODE

1. Context/Memory
2. Tool Use
3. Execution Loop
4. Evaluation
5. State Management

LLMS & CLAUDE CODE

Claude Code is not the only “coding LLM”:
1. OpenAI Codex
2. Claude Code
3. Gemini-CLI

LLMS & CLAUDE CODE

There are also a ton of IDEs that use either their own models or popular LLMs.
1. Cursor
2. Github Copilot
3. Windsurf
4. Replit
5. Antigravity (Google)
6. Opencode

CLAUDE

Let’s start using Claude.

ClAUDE

/model

CLAUDE MODELS

Opus: This is the one you should probably use
Sonnet: Sometimes ok to use
Haiku: NEVER use this for coding

CLAUDE MODELS

The bigger the model:
• The better it is at solving tasks
• The slower they are
• The more expensive they are (tokens)

CLAUDE CODE

/effort is another variable for models. This limits how many tokens a model uses per response. “high” gives the model a wide token range, “low” is much more limited. 

CLAUDE CODE

/context

CLAUDE CODE

Context is an LLM’s “memory.” It contains read content for specific files (more on this later) and your entire chat session history. For most Claude models, this is 200,000 tokens, or approximately 150,000 words. 

CLAUDE CODE

When your context reaches the maximum, it undergoes “compaction.” This reduces the memory of the model. 

CLAUDE cODE

Because of this, managing context is one of the most important things to do while LLM coding. Much of this class will be about optimizing context.

CLAUDE CODE

/clear resets your context. This is very useful, but must be used at the right moments.

CLAUDE CODE

/usage will show you how much you have used Claude in your 5-hour session limits and your weekly session limits

CLAUDE CODE

I doubt you will hit your 5-hour session limit or weekly session limit for most of this class, but if you get really into it and get agentic you might. You can pay for additional tokens in sessions, but that won’t be covered by Anthropic.

CLAUDE CODE

/resume resumes a previous chat session if you have quit CC and restarted.

claude code

/remote-control allows you to run your desktop session from your phone :)

CLAUDE CODE

/export exports your session to a file. I don’t use this often, but we might for today’s homework

Homework #1

Talk with Claude and have it develop a lesson plan for a topic you want to learn about. Focus on iteration and discussion with Claude. Publish your learnings from this process.  

HomeWORK #2

Clone a repo from Github. Use Claude to understand how the repo works, and what you can do with it.