The foundation
Most enterprise processes were designed around a screen-based human workflow: someone opens a record, reviews fields, makes a judgment, clicks a button. Bolting an agent onto a screen-based workflow produces marginal gains at best. The agent waits for the screen the same way the human did.
The leverage comes from redesigning the workflow around the agent’s actual capabilities — event-driven rather than session-driven, parallel rather than sequential, exception-routed rather than queue-based. This is process work, not technology work, and it is the work most often skipped because it is harder and slower than buying software.
What it gives you
A workflow rebuilt around the agent’s capabilities is a workflow that runs at machine cadence in the parts of the process that should, and that routes the genuinely novel back to a human with the context they need to act. Cycle time drops not by a percentage point but by an order of magnitude in the domains where the redesign holds. The work humans do in the redesigned process is the work humans should have been doing all along — applying judgment to the cases the system flagged as unusual, building the relationships the agent cannot build, designing the policy the agent will act inside next quarter.
For business leaders, the felt change is the disappearance of queue-based operations as the dominant pattern. Status meetings convened to check on inflight work get smaller. The conversation shifts from what is in the queue this week to what did the system decide, what did it cost, where did it escalate.
Where it sits
Workflow redesign is third in sequence because it depends on the first two. Without trustworthy data, an event-driven workflow propagates errors faster. Without encoded decision rights, an exception-routed workflow has no defensible logic for which exception goes to which human.
In the five-level autonomy model, the workflow foundation is the foundation that converts a Level 3 capability into Level 4 reach. A capability without a workflow built for it stays in the pilot domain. A workflow built around the capability is what scales the model — and is what the third stage of the practical sequence is actually about.