My third year leading AI coding sessions at ITP Camp

Tomorrow I'll be leading an AI coding session at NYU ITP Camp. I wasn't planning to run one this year, but during the first few days of camp I found myself having the same conversation over and over again: things get messy fast when you let the vibes completely take over a codebase.

Looking back at the AI coding sessions I've led at ITP Camp over the past three years, there's a clear progression:

• 2024: Making your first API call to an LLM
• 2025: Prompting an app into existence in 20 minutes or less
• 2026: Taming coding agent chaos in larger projects

Each year has reflected where this emerging technology was, and also where people were getting stuck: the front door, a real use case, a sustainable way to use it.

I'm really excited about this year's session because there are significant gains available from relatively low-effort practices. You don't necessarily need a new model, a new tool, or a new workflow. Often you just need a few guardrails at key points in your existing setup.

We'll be talking about topics like:

• Testing, validating, and linting
• Detecting dead code
• Keeping files and functions from growing without bound
• Defining and enforcing project rules and structure

I'll share some of the techniques that have been working for me, and I'm hoping a few attendees will share theirs as well.

The tooling keeps getting better. The challenge is making sure our projects don't get stuck with last month's way of doing things.