How is AI like a pilot lost above the clouds near Redmond, Washington?
There’s an old joke about a helicopter pilot who becomes lost above the clouds. His navigation and communications equipment have failed. Eventually, through a break in the clouds, he spots a tall office office building.
He flies close and holds up a sign:
WHERE AM I?
People inside confer, then hold up their answer:
YOU ARE IN A HELICOPTER!
The pilot immediately turns toward an airport and lands safely.
His passenger is astonished.
“How did that answer possibly tell you where we were?”
The pilot replies:
“The answer was technically correct—but completely useless. So I knew that it had to come from Microsoft.”
Funny.
But isn’t that sometimes how we use AI today?
We give an LLM an instruction. It gives us an intelligent, polished, technically correct answer—without ever determining where we’re actually trying to go.
Highly proficient AI users know how to compensate. They write thoughtful prompts, provide relevant context, challenge assumptions, recognize misunderstandings and keep steering the conversation toward a useful result.
But most people aren’t highly proficient AI users. Nor should they have to be.
They’re executives, accountants, engineers, salespeople, designers, managers and other domain experts. Their expertise is in what they do—not in prompting an LLM.
Which brings us to Henry Ford.
Ford didn’t make the automobile practical by teaching millions of people to become automobile experts. He made an automobile millions of ordinary people could use.
The Model T democratized the automobile. What if we could do the same for AI?
Instead of teaching everyone how to prompt the AI, what if we shifted that responsibility to the AI?
The user shouldn’t need to know how to prompt the AI.
The AI should know how to prompt the user.
That’s the premise behind Collaborator™.
I built a set of additional system instructions designed to make AI take responsibility for discovering the user’s desired outcome—and the information it needs to help achieve it.
Rather than immediately executing an incomplete request, Collaborator™ can ask purposeful questions, uncover missing context, surface assumptions and draw on the user’s own knowledge, judgment and experience.
No new model. No fine-tuning.
Then I built an interactive head-to-head experiment:
ChatGPT 5.6 vs. ChatGPT 5.6 + Collaborator™
Same model. Same initial prompt.
Choose one of the examples or bring your own prompt and judge the results yourself.
Maybe we’ve been looking for AI ROI in all the wrong places.