AI
LLMs, agents, and the practical edges of what they can actually do.
More than hype — AI is changing how software is built, operated, and used. I focus on practical integration: where LLMs genuinely add leverage, where they fail silently, and how to build guardrails that survive production.
Key principles
- Treat models as fallible collaborators, not oracles
- Evals before prompts before models
- Cost and latency are product decisions, not afterthoughts
Tools & tech
ClaudeAnthropic SDKOpenAI APILangChainLlamaIndex