Skip to course content
Free agentic AI fundamentals course

Introduction to Agentic AI and Workflow Automation

Unit 02.00: A goal you can tell has been reached

An agent with an unfinishable goal does not fail. It keeps going, because nothing in its instructions describes being done.

Name the artefact or decision that means finished

That is the test. "Handle the customer's request" has no end state; "decide whether INV-1187 is inside the 30-day window" has exactly one.

The table below sorts five goals.

goal                                                 done-able?
handle the customer's request                        NO   no end state
improve the response quality                         NO   never finished
decide whether INV-1187 is inside the 30-day payment window yes  one decision
produce a list of overdue invoices for June, as a table yes  one artefact
be thorough                                          NO   a disposition

Test: name the artefact or decision that means this goal is finished.

"Be thorough" is the one worth sitting with, because it reads like a quality bar and functions as an instruction to continue. An agent given it has no way to know when it has been thorough enough.

The two finishable goals share a shape: one decision or one named artefact, over identified inputs. That shape is what lets both the agent and a reviewer agree the work is over.

The mistake this prevents

The mistake is writing the business objective as the agent's goal. The business wants invoices paid on time; the agent needs to decide whether one specific invoice is overdue. Keep the objective in the design document.

Takeaway

A goal must name the artefact or decision that means it is finished. Unfinishable goals produce agents that elaborate until a ceiling stops them, which is a cost problem before it is a quality one.