Notes on building.
Essays on AI-native engineering, agentic systems, and the architecture of modern developer tools. Written between shipping.
Pair programmer or delegate? Two ways to work with AI
Most arguments about AI coding tools confuse two different modes. When to hold the keyboard, when to review the diff, and how the task decides.
Read the essayEvals for code agents: what to measure when the demo lies
Every code agent demos well. The metrics that actually predict production: merge-without-edit rate, per-lane autonomy, and evals built from your own git history.
Read the essayHow we let AI write Laravel migrations without losing sleep
Schema changes are the scariest thing to hand an agent. The four mechanical checks that make AI-written Laravel migrations boring and safe.
Read the essayContext engineering is the new system design
When a model does the work, what it can see is the system. Why designing context became the highest-leverage architecture work I do.
Read the essayThe Model Context Protocol, in practice
MCP is turning into the USB-C of AI tooling. Lessons from building servers: tool design, error messages as prompts, and context budgets.
Read the essayDesigning agentic systems that don't go off the rails
Agents rarely fail loudly. They produce work that looks right and isn't. The guardrail layers that hold up in production, and the ones that don't.
Read the essayWhy software development is becoming AI-first
The slow part of shipping software is no longer writing the code. Notes on what that changes for engineering teams.
Read the essay- 01
The economics of AI-generated code
Deep dive - 02
Legacy codebases are the real frontier for AI agents
Essay - 03
What a solution architect does when AI writes the code
Essay - 04
Multi-agent orchestration: when one agent isn't enough
Deep dive - 05
The infrastructure behind an AI development platform
Deep dive - 06
Open-sourcing AI agent skills: what I learned shipping them
Engineering
Publishing these one at a time. Want one prioritized? Tell me.