Tag
Discusses a common failure mode in AI agents where the model claims to have executed a tool call without actually firing it, and advocates for trusting execution receipts over agent narration to ensure reliability in production.
Engineering notes comparing three approaches to unifying access to multiple LLM providers (OpenAI, Anthropic, Google) behind a single internal interface, discussing trade-offs in API normalization, native SDK usage, and gateway patterns.
The article discusses common failure patterns in agentic AI systems, specifically 'dumb AI loops,' citing issues like state poisoning and data leaks observed in Claude Code deployments.
This post outlines a comprehensive 9-layer AI production architecture, emphasizing components like RAG pipelines, security guards, observability, and evaluation to distinguish robust production systems from simple demos.