How far out of its own comfort zone can you push a coding agent? Is it possible to turn them into a filing assistant that can work through 15 years of digital clutter and handwritten notes?

In this talk, I’ll take you on a wild ride from a basic CLAUDE.md into a system that helped me sort and migrate a long-neglected personal archive. We’ll walk through different ways of configuring Claude Code - including building memory, creating specialized tools, attempting to add guardrails, and much more. Along the way, I’ll cover what worked, what broke, and how to keep work efficient without burning a small rainforest of tokens.

While this is a talk about a specific Claude Code use case that was not advertised on the metaphorical box, the message is about something deeper. It’s about more than a note cleanup; it’s about how we can use coding agents in non-obvious ways, how technical leaders can fit them into their workflows (beyond code), and why the structure of our personal and institutional knowledge may need to change for the AI era.

You’ll get both practical tips on how to (mis)use Claude Code and a chance to challenge yourself to find good (and bad) use cases for the latest hyped technology.