Define the job.
Agree the objective, inputs, permitted tools and completion criteria. Decide which decisions remain with a person.
AI agents & automation
Give the agent a job.Engineer the workflow.
Agents that retrieve information, use tools and carry out multi-step work. Built around a bounded task, explicit permissions and a way to check the result.
Explore the capabilityENGINEERING FOCUS
01Retrieve
02Decide within limits
03Act and verify
Autonomy is a design decision.
A predictable workflow may need very little autonomy. More open-ended work needs tighter evaluation, execution limits and supervision. Start with what the task actually requires.
Connect the work.
Integrate data and applications. Design state, validation and retries around the workflow, not around a chat transcript.
Test the boundaries.
Exercise denied access, contradictory evidence, tool errors and interrupted work. Measure completion and the quality of the action.
A possible workflowIllustrative example.
Prepare an operational request from approved information, validate the fields and pause for authorisation.
Use the least autonomy needed. More agents, more tools or longer prompts do not automatically produce a better system.
A closer look
Good questions.
Straight answers.
Can we begin with one task?
Yes. A bounded, measurable workflow makes it easier to assess value, risks and integration effort.
Will people remain involved?
Where the task requires it. Approval, exception handling and operational ownership should be defined before the system is used.
Technical reference: Anthropic: evaluating AI agents (opens in a new tab)
Keep exploring