Blog
Observations on building agentic systems: what works, what fails, and why the harness matters more than the model.
Read for orientation, then move to services when you are ready to pressure-test a real workflow.
Arrange It So It Can Just Go Forever
Karpathy's auto-research beat his two decades of hand-tuning overnight. What 'arranging it' actually requires — from specification discipline to perception engineering to the verification stack that replaces you.
Read article →The Arrow of Certainty: On Agents That Act Before Being Asked
What a physicist's theory of promises, a pair of organizational behaviorists, and a hippie mechanic reveal about the deepest challenge in building proactive AI agents.
Read article →The Dish and the Recipe: On Building Tools That Agents Can Grasp
What a philosopher of games, a software design heretic, and a distributed systems theorist reveal about why tool design for AI agents is really about what we leave out.
Read article →The Narrow Path: On Agents That Remember
What a philosopher of narration, a specification writer, and the deep history of feedback loops reveal about the difference between agents that accumulate data and agents whose work compounds.
Read article →The Compound Engineering Loop
Every mistake your agent makes should become a permanent improvement. Here's how to build systems that learn from every encounter with reality.
Read article →Why Most AI Agents Fail in Production
The gap between an AI demo and a production system is enormous. Here's what we've learned about closing it.
Read article →