██████╗ ██╗ ██████╗ ██████╗ ██╔══██╗██║ ██╔═══██╗██╔════╝ ██████╔╝██║ ██║ ██║██║ ███╗ ██╔══██╗██║ ██║ ██║██║ ██║ ██████╔╝███████╗╚██████╔╝╚██████╔╝ ╚═════╝ ╚══════╝ ╚═════╝ ╚═════╝
Why large language models sometimes look like they have a will. Emergence, levels of description, and why "looks like agency" ≠ "has inner will". Connecting philosophy, neuroscience, and AI safety research.
Sometimes the best project ideas come from stepping away from something you loved. How quitting serious mountaineering led to building RePeaks — a gear marketplace with AI-powered inventory and valuation features — in a single weekend.
I compare two modes of working with AI tools: "agentic coding", where AI acts like a structured co-worker with plans and checklists, and "vibe coding", where I explore ideas more freely. In the post I show how I combine both modes when building MVPs and refactoring existing codebases.
How AI assistants change the development workflow. Real examples of using Cursor for prototyping and Claude Code for refactoring.
Implementing Donald Michie's matchbox learning algorithm in TypeScript. How reinforcement learning works without neural networks.