> $ cat ~/soul.md
cd ~/blog/A working engineer's notebook, unfinished thinking, papers worth a second read, and learning shared.
Working with LLMs has moved through three phases: prompt engineering, context engineering, and now harness engineering. Here's why the environment you build around the model matters more than the model itself.
A Feynman-style deep dive into the Transformer architecture. We build intuition from the simplest possible idea up to full GPT implementations, tracing Karpathy's progression from micrograd to nanochat.
A deep dive into AI agents, starting from the simplest possible idea and building up to production-grade agentic systems with tool use, memory, reasoning loops, and multi-agent orchestration.