4

"Are you sure?"

The three most valuable words in this entire library, and the story of how they dissolved weeks of planned work in one afternoon.

Here is the most expensive sentence that was ever said to me by a machine: "That information isn't available from this system."

It came up while my AI and I were building something that needed a particular piece of data. The AI stated, flatly and helpfully, that the system we depended on did not provide it. And because the claim sounded technical and authoritative, I did what most people do: I accepted it and started planning around it.

And what a plan it was. If the data wasn't available, we would have to reconstruct it ourselves. Extra storage to track it. Extra logic to keep it fresh. Extra cases for when our reconstruction disagreed with reality, which it would, regularly, forever. The plan grew to weeks of work. It was clever. I was a little proud of it. Every piece of it was a reasonable answer to the constraint.

Then, almost as a reflex, before committing to all that work, I asked one question: "Are you sure?"

Not a rhetorical are-you-sure. A real one: go and check the actual source. Look at what the system actually returns, not what you believe it returns. The AI checked. The data was available. It had been available the entire time, one small setting away. The wall we had spent weeks designing around did not exist.

I think about that afternoon a lot. Nothing in the plan was stupid. The reasoning was airtight. That is exactly what made it dangerous: reasoning is a machine for building beautiful structures on unchecked ground. The AI had not lied to me, either. It had predicted a plausible technical claim, in its usual confident voice, and plausible was all it ever promised. The failure was mine: I let a claim become a foundation without making anyone touch it.

Since that day I have a physical rule: before building around any wall, walk up and touch it. Ask the system. Read the actual document. Call the actual endpoint, click the actual button, look at the actual data. Every "that's impossible" and "it doesn't support that" and "they don't allow it" gets one are-you-sure before it is allowed to cost me anything. You would not believe how many walls are made of fog.

Stop here. Actually sit with this before you scroll on.

Which of the things you believe about this problem have you actually checked?

Try this nowUnder 30 minutes
  1. Find a recent "impossible." Something an AI, a colleague, or your own memory told you can't be done: an app that "doesn't have" a feature, a service that "doesn't support" something, a rule that "doesn't allow" it.
  2. Ask your AI: "Are you sure? Check the actual source." Make it look, not reason. Documentation, settings, the real interface.
  3. Touch the wall yourself if you can: open the settings page, read the actual policy, try the actual action.
  4. Keep a tally. Seriously, start a note called "Walls." Each time, record: was the wall real? Most people's tally converges around half fog. Yours will change how readily you believe the next wall.

Two questions before you go

Answer, then say whether you were sure or guessing. Being honest about which is the skill being trained.