Week 7

Demo Day & Agents

NYU · Whatever Generative AI Is Doing Now · 2026

SCHEDULE

Week 1 Install and Claude Basics
Week 2 Research and Writing with Claude
Week 3 Chrome Use, Computer Use & MCPs
Week 4 Build Your Own Software, Part 1
Week 5 Build Your Own Software, Part 2: AI Generation Tools
Week 6 Deploy Your Site
Week 7 Demo Day & Agents

Section 1

Demo Day

Section 2

Building Agents Into Your Product

Maybe you’ve been thinking...
Why not just run Claude in my app?

Flora's Fauna interface showing a Make anything prompt with suggested actions

florafauna.ai

Anthropic announcement: Claude for Creative Work

anthropic.com/news/claude-for-creative-work

Scene Ripper app showing detected scenes and an agent chat panel

EVERY USEr ACTION IS AN AGENT ACTION

To create an agent in your app, you need to give it the same tools as a human user.

WHAT THAT MEANS

Every button in your app is doing something underneath — saving a drawing, generating an image, changing the brush.

Build a Save button? Add a save_drawing tool. Build a Generate button? Add a generate_image tool. The user clicks. The agent calls. Same result.

EXAMPLE: YOUR DRAWING APP

Things your user can do:

• Save a drawing
• Generate an image from the canvas
• Change the brush size
• Clear the canvas
• Share by URL

Give the agent a tool for each one.

EXAMPLE: YOUR DRAWING APP

Once the agent has all of those, a user can ask for things you didn’t put a button on:

“Generate three variations and save the best one.”
“Auto-save every 30 seconds.”
“Make a square version too.”


You didn’t build those features. The agent composed them from your tools.

HOW TO BUILD FOR IT

• Keep tools small and named like verbssave_drawing, not handle_save_request
• Tell the agent what outcome you want, not what to click

WHERE THIS COMES FROM

This idea comes from the team at Every — the same group behind the /ce: commands you’ve been using.

What we just covered (parity) is the first piece of a longer framework.

EVERY’S GUIDE

every.to/guides/agent-native

It covers the principles, common patterns, and how agent-native apps get better over time without shipping new code.

USE IT IN CLAUDE CODE

The framework is also installed in your Claude Code as a skill. To pull it into a session, just ask:

“Use the agent-native-architecture skill to help me think through how to make my app agent-native.”

It’ll walk through the framework with you and suggest specific changes for your app.

ONE MORE: /CE:COMPOUND

When you build something agent-native, you’ll solve problems no one’s solved before.

After each one, run /ce:compound. It writes up what you fixed — symptom, cause, fix — as a reusable skill in your repo.

The bug you fixed today fixes itself next month.

These tools will keep changing. The process 
describe what you want, then check the work won’t.

THANK YOU! STAY IN TOUCH

• Email: dvsmethid@gmail.com, ds6671@nyu.edu
• Site: artificial-images.com