← Back to articlesAgent basics

Why is an agent's plan not proof of completion?

A plan states intended actions; a receipt records what happened.

The short answer

A file, a passing command and a reachable page prove different things. Request outputs, checks and incomplete items rather than accepting done as evidence.

Try it in three steps

  • Write an observable completion criterion.
  • Request the artifact location and check record.
  • Open the critical result yourself.

A practice prompt

Before you start

Classify three fictional progress messages without running tools.

Practice material (fictional)

A: Planning an inventory. B: Wrote inventory.txt. C: Read it back; two lines match the folder, zero skipped.

Replace placeholders in the prompt with this material, or append the material if there are no placeholders. Reading time excludes doing the exercise.

Report actual changes, output locations, checks and incomplete work separately. Do not mark unexecuted steps complete.

What to look for (not a recorded test result)

A is a plan, B reports a write, C provides checking details. Inspect the actual file too; self-report is not independent evidence.

If it goes wrong

If a screenshot is simulated, request the real artifact or preserve an unverified status.

Check your result

Another person can inspect the result using artifacts and records rather than trusting a statement.

References and scope

The links below support the concepts or product navigation. The steps and exercises are our teaching suggestions, not official certification or measured customer outcomes. Product interfaces and availability may change. References checked on 2026-10-17.

Start with one task

Describe the current process and its slowest step. The initial conversation is free.

Let’s talk ↗
Need help? See small-scale implementation services ↗

Keep reading