Stop asking AI to finish. Tell it what finished means (Part 3 of 4)

Stop asking AI to finish. Tell it what finished means (Part 3 of 4)

Natalie Lambert
Natalie LambertFounder, GenEdge
November 17, 2026
5 min read

Welcome to Prompt, Tinker, Innovate—my AI playground. Each edition gives you a hands-on experiment that shows how AI can sharpen your thinking, streamline your process, and power up your creative work.

Part 3 of 4: Get AI to finish the job. Plan → Loop → Goal → Schedule. This week: goal.

This week's playground: A finish line that stays active

Last week, you asked AI to check a rough draft, fix what it could, and check again within one response. The built-in Goal feature runs another kind of loop: it keeps the completion condition attached across turns. If AI tries to finish before meeting it, the goal can keep it working or surface a question for you.

Here is the practical difference:

  • Prompt-driven loop (Part 2): In one response, AI checked a rough draft, fixed what it could, checked again, and reported anything unresolved.
  • Goal-based loop (this week): The goal stays active while you and AI go back and forth.

For a short, straightforward draft, a prompt-driven loop is often enough. A goal-based loop is best when the work may need another attempt or a human answer before it can be called complete.

Why this matters

A disputed date cannot be fixed by polishing the wording. The prompt-driven loop can point it out in one answer. With the Goal feature, the unresolved requirement stays attached while you answer and AI revises.

You do not have to restate every requirement after each turn. The Goal keeps the finish line in view, though you still need to review the result.

Use case spotlight: The same update, with a persistent finish line

Use the same two notes and checklist from Part 2. One note says training starts December 3; the other says December 10. Watch whether the tool keeps that conflict open, asks you to resolve it, and checks the revised update after you answer.

In both tools, the practical benefit is the same: the finish line stays active across turns. If the AI asks which training date is right, you can answer without repeating the checklist. The difference is in how the tools manage the goal:

  • In ChatGPT Work on desktop, /goal shows your active goal and lets you pause, change, or clear it.
  • In Claude Code, a separate AI checks the goal when Claude tries to stop and sends Claude back to work if it has not been met.

Neither check can tell which date is correct from conflicting notes. You supply that fact and review the final update.

Where this works

  • ChatGPT desktop app: Select Work, then type /goal and paste the prompt. Check the command menu by typing / if you don't see it.
  • Claude: /goal lives in Claude Code. Click the </> tab at the top of your Claude window, type /goal, then paste the prompt.
  • Standard Chat: Use the prompt below with the extra line immediately after it. Check the result yourself; standard Chat does not enforce a goal.

Your AI experiment: Try this prompt

👉 Time to tinker: The same fictional notes from Part 2 are included below, so you can start here. In the ChatGPT desktop app, select Work and type /goal followed by the prompt. In Claude Code, type /goal followed by the prompt. Unlike last week's one-response test, the built-in Goal feature keeps the checklist active across turns. AI should spot the conflicting dates, ask you which is right, and check the final update against the same checklist after you answer.

📝 Prompt (built-in Goal feature): Type "/goal" first, then paste this prompt.

Create a team-ready project update from these fictional notes. If the notes conflict, ask me which fact is correct and wait for my answer before calling the update complete. Operations note: The new-hire training guide draft is ready. Jordan owns the guide. Training starts December 3. The FAQ still needs legal approval. HR note: The screenshots are updated. Training starts December 10. The FAQ still needs legal approval. The goal is met only when the update passes every item: - Under 150 words. - Four labeled sections: Status, Progress, Next step, Risks. - Use only facts from the notes and my answers; do not invent an owner, date, or approval. - Confirm the training date before calling the update complete. - Show any fact that still needs a person's answer as [needs input].

📝 Prompt: If you are using standard Chat, add this line to the prompt above:

If you need a fact from me, ask and stop. When I answer, revise the update and check every item on the original list again before calling it complete.

💡 Pro tips: Write a finish line worth enforcing

  1. Put the checklist in the goal itself. The goal text is what gets checked, so a requirement you add later in the chat may not count.
  2. Tell it when to stop and ask you. A disputed date needs a person's decision, not a confident guess.
  3. Bound the task. If a requirement cannot be met, ask for the missing fact rather than trying indefinitely.
  4. When AI asks for the date, answer and watch whether it revises without making you restate the checklist.
  5. Still read the update. The tool checks your list; whether the status reflects reality is still your call.

Make it yours

What would you change to make this useful in your own work?

Until next time—keep tinkering, keep prompting, keep innovating.

Next week in Part 4: put a useful information watch on a schedule.

📩 Not subscribed yet? Hit the button at the top. You won't want to miss what's next.