Articles
Field notes on building AI that does real work: the loops, the context layers, the observability, and the systems agents act on.
- 4 min read
Enterprise Context Layers for Agents: From Semantic Models to Reasoning Graphs
Most enterprise AI projects fail because the agent reads the company's documents without understanding the company. The fix is not a bigger model. It is a context layer.
RAGKnowledge GraphsContext - 3 min read
Shopping Agents Are Here. The Catalog Is the New Interface.
Buying is moving into the chat window. The protocols decide how money moves, but the catalog decides whether your product was ever in the conversation. A look at both sides of agentic commerce.
EcommerceAgentsCatalog - 3 min read
Putting Agents to Work Against the Systems Your Business Runs On
Most agent demos act on toys. The interesting work is acting on your ERP, where the orders and the ledger live. Here is how to let something non-deterministic touch a system of record safely.
ERPCloudEnterprise - 4 min read
You Cannot Manage What You Cannot See: Observability and Cost Control for AI Agents
Most teams ship an agent they cannot see inside. The fix is not a bigger model, it is instrumentation: spans, per-request cost, and the checks that catch a silent failure before a customer does.
ObservabilityCostProduction - 4 min read
Loop Engineering: The Part of Agents That Actually Ships
The model is almost never the thing that breaks in production. The loop is. A look at the boring control plane that separates a demo from something you can put in front of a customer.
AgentsLLMReliability