Agent Basics: from curl to tool calls
The first episode of Agent Basics. No frameworks, just HTTP requests to a local model, structured output and a tool call executed in Python.
Blog
Essays on software engineering, data, and building with AI.
My Ten Rules of Agentic Engineering
Ten rules for building software with AI agents. Most are about the system around the agent, not the agent itself.
The 45 Minute 'Long-Lived' Feature Branch
How AI-assisted delivery makes shared context and trunk-based development more important, not less.
Sakana AI's Fugu: Is this where model routing should live?
Fugu shows that learned model routing can work, but application teams need routing they can control, observe and evaluate.
Knowledge Products
If you want AI to do knowledge work, then you need to build knowledge products.
Does every feature we build with AI need a token budget?
Treat AI features as products with explicit build, run and maintenance budgets tied to a hypothesis about their value.
From API-First to API-Complete
AI agents need every organisational capability to be addressable through a typed interface with a stable contract.
Hardening Codebases for Agentic Coding: 10 Essential Techniques
Ten practical techniques that make codebases safer and more effective for AI coding agents.
Building your own CLI Coding Agent with Pydantic-AI
Build a coding agent from open source tools and tailor it to your own engineering environment and development standards.