“Taxonomy for Agent Systems (T4AS)” introduces a simple but powerful way to tame today’s chaotic landscape of AI agents. Steve Vitka presents T4AS as a reference architecture that cleanly separates three roles: the Agent (“strategist” that plans but never acts), the Workflow (“general” that orchestrates tools), and the Workspace (“battlefield” where certified tools and APIs actually run). By enforcing these boundaries, T4AS turns ad‑hoc agent stacks into secure, auditable, and composable systems rather than brittle prompt‑spaghetti.
This session will walk through real examples of how the triad stops prompt‑injection and runaway tool use—showing what the Agent requested, what the Workflow chose to execute, and what the Workspace actually did. Vitka will also connect T4AS to emerging efforts around loyal personal agents, verifiable workflows, and digital twins, outlining how a common taxonomy can help the ecosystem converge on interoperable, people‑first agent standards.
The presentation will take place in Ballroom F on Thursday, March 5, 2026 - 16:00 to 16:50
