Automation is the wrong ambition.

Most AI automation pitches sound the same. Find a repetitive task. Point a model at it. Save a few hours a week. Multiply by headcount. Present the number.

It works, in the narrow sense. It also leaves the organisation exactly as it was, only slightly faster. The same handoffs, the same queues, the same people waiting for the same approvals. The task got automated. The work did not change.

Work is a flow, not a list of tasks

Inside any organisation, work moves: from a customer to a system, from a system to a person, from a person to another system, and back. Most of the time is spent in the moves, not in the tasks. Waiting for context. Re-keying data. Asking someone who knows. Escalating.

Agents and event-driven systems are interesting because they can operate in the moves. An order arrives and the system already knows the customer's history, checks stock, drafts the exception, and asks a person only when a judgement call is needed. That is not a faster version of the old process. It is a different process.

What redesign looks like in practice

• Model the work as events and decisions, not as departments and tickets.
• Keep deterministic steps deterministic. Agents handle judgement, not arithmetic.
• Put people at the decisions that carry weight, and give them the full context when they get there.
• Make everything observable. A system that acts at night has to explain itself in the morning.
• Measure throughput, quality and time-to-resolution, not hours saved.

This is harder than automation. It requires understanding the organisation, integrating with systems that were never designed to talk to each other, and designing interfaces for the moments where people step in. It is also where the leverage is.

Automate a task and you save an hour. Redesign the flow and you change what the organisation can do.

If a proposal ends in hours saved, ask what happens to the hours. If nobody knows, the ambition was too small.