What I'm focused on.

Building

• Production-grade AI applications — agents and retrieval that hold up past the demo
• Scalable backend platforms and cloud-native systems
• Developer experience and platform reliability — the leverage that compounds across a team

Exploring

• Agentic and multi-agent systems — where coordination beats a single sharp model, and where it just adds latency
• Evaluating LLM systems — turning “seems better” into something you can measure
• Context engineering and retrieval beyond naive RAG
• Durable, long-running AI workflows — state and orchestration that survive failure
• The economics of production AI — latency and cost as first-class constraints

Thinking

• Simplicity scales better than clever abstraction
• Most AI products die after the demo — the model was never the hard part
• Reliability before scale; observability before both
• Premature complexity is the most expensive debt
• Engineering trade-offs age better than engineering trends

Reading

• Designing Data-Intensive Applications — the systems classic, worth a re-read
• Papers on agent evaluation, planning, and long-running tool use
• How the frontier labs frame context, memory, and reliability