Ruby Melbourne: June 2026

A bit of a recap on what I've learnt from using Ruby to interact with Sonos speakers and music service APIs.

Your cleverest code is probably overfitting. A simple rule that ignores almost everything routinely beats a sophisticated model that uses all the data — and it's not a fluke, it's a theorem. We'll trace the same maths that explains overfitting in machine learning straight back to how minds make good decisions, and figure out what it says about every premature abstraction you've ever written. A talk about why throwing information away is sometimes the smartest thing you can do.

"The dream loop was too long."

Most of us have used AI to write code. This talk is about what happens when you go further — when AI stops being a coding assistant and starts being the orchestrator of your life.

Covering a journey from:

• Scout documents — badge writeups, risk assessments, hike plans (boring but effective)
• Code — mass refactoring, security analysis, decoupling (yes yes, we've all done it)
• Electronics — where it gets weird. AI directing hardware projects, component purchases at 2am, and backyard chemistry experiments that got slightly out of hand
• Music composition — AI writes a four-part recorder arrangement; you negotiate with your children to perform it

The thesis: you are the agent, AI is the master controller. The dream-to-execution loop has collapsed. The question is whether you're comfortable being the one who has to physically show up.

Live demo: a short Ruby script (AI-written, naturally) feeds this very talk into Gemini's video API and returns key takeaways, best moments, and a summary for anyone who fell asleep.