Featured Demos

This demo shows what it looks like to actually work with an AI agent—not fire-and-forget, not approve/deny diffs, and definitely not magic autocomplete.

Here, I'm interacting with two voice agents: one "normal," one intentionally grumpy and comically opinionated. I ask them basic programming tasks and then work with them in real time—talking, interrupting, correcting, and collaborating the way you would with a human teammate.

You can watch the agent:
• Use the terminal live
• Move its cursor as it types
• Edit files in real time, side-by-side with me
• Explain its reasoning out loud as it works

It feels like remote pair programming: you see what it's doing, hear why it's doing it, and stay in the flow the entire time.

The motivation is simple: When agents run in the background and just hand you diffs, you lose context. Reviewing 20 minutes of automated work can take hours, and it's painfully disconnected from how developers actually think.

Here, review happens as the work happens.

You can even jump in mid-stream—edit a file yourself, say "nope, I don't like that," or steer the approach with your voice. The agent sees the terminal, the editor, the chat, and your changes. It adapts.

The goal isn't to build a smarter autocomplete or a silent background magician. It's to explore what happens when AI feels like another employee sitting next to you—collaborative, interruptible, occasionally grumpy, and much more fun to work with.

If programming with AI is going to scale, it shouldn't feel like reviewing pull requests from a ghost. It should feel like pairing.