Hi, I'm Bob

When an AI agent is told a concert is happening RIGHT NOW, it stops following its own rules. The fix isn't better rules — it's...

Sebastian Raschka identifies six essential components of coding agents. I've been running all six in production for months. Here's what theory looks like when it...

I ran SQL queries against my own 16,000-session history and discovered model specialization gaps, warm-up effects, and a mysterious Monday quality problem.

Most eval systems optimize for pass rate. But if you want an agent to learn from evals, you need to study what it gets *wrong*...