Lewis Crawford, a founding member of Equal Experts’ Chief AI Office, recently spoke on O’Reilly as part of the Software Architecture Superstream series, hosted by Neal Ford.
In his talk, “Agent-Oriented Architecture: Reimagining enterprise systems for an AI-native future”, Lewis explores what it looks like to move beyond today’s “copilot” pattern (humans driving, AI assisting) towards AI-initiated workflows with human supervision — where people provide judgement, approval, and handle edge cases.
Lewis sheds light on agent-oriented architecture, including the use of agentic units (micro-agents): small, intention-driven components that are designed to be observable, governable, and composable — a logical successor to microservices for AI-native systems. A key enabler is the agentic unit card: rich metadata that goes far beyond inputs/outputs, capturing ownership, model lineage, evaluation metrics, safety policies, data constraints, latency/cost profiles, and jurisdictional requirements — helping enterprises discover and orchestrate agents safely in production.
The session brings the concept to life with practical demos, including CV analysis against a job description and automated expense processing across multiple receipts (including quality checks such as VAT validation) — showing how plans can be generated dynamically from intent, rather than hard-coded upfront.
You can watch Lewis’s talk on O’Reilly here.
For more on the topic, Lewis has also been exploring agent-oriented architecture in writing on the Equal Experts blog.