Working well with the Agent

Practical habits for getting dependable operational work from the Agent.

Good Agent use is less about prompt tricks and more about clear outcomes, trustworthy evidence, explicit boundaries and reusable corrections.

Start with the business outcome

Describe the job in operational language. Tell the Agent what should be true when the work is finished rather than scripting every tool call.

Make source authority clear

If sources can disagree, say which one wins.

Examples:

• ERP owns the current order quantity.

• The signed policy outranks an old presentation.

• Customer-specific rules outrank generic industry guidance.

• A remembered preference does not outrank a current system record.

Ask for evidence

For important work, ask the Agent to make the basis of its answer inspectable.

Useful evidence can include the records it checked, the document supporting a value, the calculation behind a recommendation, the file it produced or the system change that now exists.

A fluent answer is not itself proof.

Make the human boundary explicit

Say what the Agent may decide, what it may prepare and what still needs a person.

If the Agent reaches a point that genuinely needs judgement, let it ask rather than guess. Some boundaries can also be enforced through approvals or Workflow Requests.

Give useful corrections

When something is wrong, include:

• what is wrong;

• where it appears;

• which source or example should have been used;

• what you expected instead.

For generated documents, inspect the document itself before it leaves the organisation.

Keep exceptions separate from normal rules

If this case should intentionally differ from the normal method, say so.

If the same correction keeps returning, move it to the narrowest durable layer that owns it—for example the Skill, maintained data, Knowledge, Memory, Project instructions, Workflow or approval setup.

Use Skills for recurring methods

A Skill is the right place for a reusable way of doing work without turning the whole task into a rigid sequence.

Let the Agent use available Skills when they fit. Invoke one deliberately when you need a specific method followed.

Use Workflows for stable background paths

When the route becomes sufficiently known and repeatability matters more than interpretation, move that part into a Workflow.

A common pattern is:

Agent investigation → Workflow execution → human decision if needed → Workflow continues → result back to the operator

Improve from real use

Response feedback, session feedback, repeated corrections, failed cases and recurring manual intervention are useful evidence.

Use that evidence to improve the maintained setup. Do not turn every bad result into another sentence in a giant prompt.

When something goes wrong

Diagnose the layer before changing the setup or retrying repeatedly.

• Wrong answer or output → inspect the source evidence, activity and result.

• System action unavailable → check Project scope, Connection and permissions.

• Same mistake recurring → improve the durable method or source.

• Failed response → use Retry when it is safe to repeat the Agent response.

For deeper investigation, see Troubleshooting and support.

For consequential work, verify the resulting state in the system or maintained source that owns it.