Thoughts on AI, Coding & Automation

Taste, distribution, and iteration speed: how AI changes product building and the role of development.

UI automation can unblock legacy systems fast, but it has tradeoffs. Here’s when it makes sense and how to keep it reliable.

Why “just give the agent a token” is risky. Least privilege, audit trails, approvals, and safe tool contracts.

Memory, tools, state, evals, and observability: what changes when you move from chat to execution.

Why chat is not a system. Contracts, state, tool governance, and verifiers.

How to turn agents from black boxes into manageable systems: tracing, metrics, auditing, and quality control.

Chunking, metadata, reranking, and quality control: the practical levers behind reliable retrieval.

Stop selling models and bots. Start shipping outcomes with measurable KPIs and clear contracts.

A simple pattern to keep LLM automation under control: routing, narrow specialists, and quality checks.

Roles, memory, tools, governance, and the patterns that turn a demo agent into a production system.

How to set up and use Claude Code for efficient development — from basic commands to advanced workflows.