Thoughts on software engineering
-
My 2026 guide for becoming an AI-native developer
A practical 2026 guide to becoming an AI-native developer with coding agents, covering context, harness engineering, software factories, feedback loops, and validation.
-
Annotations for Anthropic’s AI-Native SDLC Playbook
A critical look at Anthropic’s AI-Native SDLC Playbook: what works, what doesn’t, and where its advice on agents, design, evals, and reviews falls short.
-
Harness Engineering 101: How to go from zero harness to a harness that survives model upgrades
My advice on how to add your first piece of coding agent harness to your codebase and how to approach its evolution.
-
Harness as Dependency, and the messy workspaces of harness engineers
A practical exploration of managing reusable “harnesses” across local and cloud environments — and why we still lack good patterns for doing it well.
-
From local to global, and other ways of dealing with instruction bloat and bike shedding
A practical guide to shared agent instructions — what to include, how to avoid instruction bloat and bike shedding, and simple ways teams can evaluate whether their instructions actually improve results.
Patterns for coding agents
-
Planning with Gherkin
Use Gherkin during planning to help coding agents write fewer, higher-value integration, system, and E2E tests with a simple human-in-the-loop workflow.
-
Navigator Agent
Learn the Navigator Agent pattern: a pair-programming workflow where AI guides implementation while you write the code, build deeper understanding, and sharpen your technical judgment.
-
Human Todos
Human Todos is a practical pattern for adding human review checkpoints to coding agent workflows, helping teams validate AI-generated code at scale.