Notes on applied AI.
Articles on integration architecture, AI agents and process automation, written from the experience of putting them into production.
MCP-Led: semantic integration for enterprise systems
Integration paradigms have always treated the problem as a syntactic one: moving data from A to B while honouring a technical contract. This pattern frames it as a semantic problem, and defines an architecture in which an agent understands the integration, executes it, and then generates the automation that will repeat it.
Your business will need an agent, just as it once needed a website
For thirty years the digital storefront was designed to be looked at by a person. That is changing: more and more decisions are made by an agent acting on someone's behalf. And the next step isn't agents serving people — it's agents negotiating with each other.
Your integration architecture is not ready for what's coming
You've spent years designing flows, defining sequences, coding field-by-field transformations. And it works. But the consumer of your integrations is about to change: it's no longer an application that executes what you tell it, it's an AI agent that reasons, decides, and acts. The fundamental unit is no longer the flow — it's the tool.