AI agents

What it is

In this Primer, an AI agent uses a model to choose steps and tools toward a goal. A workflow follows a sequence defined in advance; an agent can decide its next action as it receives new information. This follows Anthropic: workflows and agents; terminology varies across the industry.

Why it can add value

Some work requires a changing sequence: inspect a case, identify missing information, consult a permitted source and decide which allowed action should follow. Dynamic coordination may be useful when a fixed route becomes cumbersome. It also creates more behavior to evaluate.

When to investigate it

• The task genuinely benefits from selecting the next action dynamically.
• Tools, information and permissions can be bounded to the task.
• Completion can be recognized and consequential actions have appropriate checks.
• Failures can be detected, contained, escalated and learned from.

The operating-model connection

Delegating an action changes who or what may act, when a person must intervene and who remains accountable. Define the scope, owner, allowed actions and recovery path together. Reliable knowledge and clear decision rights matter as much as access to a model.

In customer onboarding, an agent might coordinate permitted follow-up actions. It would still encounter disagreements about what complete means. Use the changed-condition exercise to test whether the proposed scope matches the actual constraint.

A useful next question

What useful choice must the system make that a defined workflow cannot reasonably handle, and what evidence would justify delegating it? Compare the smallest worthwhile scope with AI assistance and rules-based automation.

Continue with the Operating Blueprint to connect delegation, knowledge, people and control. Use Measurement to make the evaluation concrete.