Don't build an agent. Redesign the work.

Every week someone asks me: "How many AI agents should we build?"

My answer is to start with fewer.

Automating a process doesn't improve it

Most agent projects start with a task someone dislikes: triaging tickets, drafting replies, reconciling records. An agent is built to do that task exactly as it is done today.

The result is usually the same process, run faster. Its handoffs, unclear approvals and rework tend to remain, and any issues it had now occur more quickly.

Start from the outcome

The organisations getting real value from AI work the other way round:

1. Define the outcome. What should be true for the customer, the patient or the shipment when the work is done?
2. Redesign the steps. Which steps need human judgement? Which need speed? Which shouldn't exist at all?
3. Place the AI. Only then decide where an agent belongs, and what it may do alone.

Often the answer is fewer agents, doing narrower jobs, inside a process that was redesigned around them.

The future of AI is better-designed work

The greatest value is unlikely to come from the number of agents an organisation deploys. It comes from rethinking the work, with agents as one part of a better design.

Simrandeep Sethi is CEO of 2SD Technologies. This is part one of our leadership series on AI agents.