Stop Asking Which Jobs AI Replaces
The wrong question has been running boardroom conversations for two years: which roles does AI eliminate? It's the wrong cut because it frames AI as a headcount event rather than a system-design problem.
Paul Roetzer's MAICON 2026 keynote offers a sharper frame. He proposes three named roles that map to how human-agent workflows actually function:
- Architect — designs the AI-native system: which agents run, what they're authorized to do, how they connect to data and tools.
- Orchestrator — directs humans and machines toward a business outcome in real time. This is the supervisor role most orgs are currently running without a job title.
- Apprentice — learns judgment in an environment where traditional entry-level tasks have been delegated to agents. The development path has changed; the need for human judgment hasn't.
The value of named roles isn't organizational aesthetics. It's accountability. When something goes wrong in an agentic workflow — a message sent to the wrong segment, a CRM record overwritten, a pricing quote issued incorrectly — someone needs to own the failure. Without explicit role definitions, ownership diffuses instantly.
The Governance Warning Buried in the Model
Roetzer's sharpest point isn't the taxonomy. It's the threshold it implies: the moment an agent can touch systems of record, send external messages, or make customer-visible changes, you have crossed from running a productivity experiment to running a governance risk.
Most teams haven't registered that shift. They're still framing agent deployments as