Answers Are Easy. Judgment Is Knowing Which Answer Is Right.

We've spent years teaching AI to answer us. The next skill is teaching it to judge — and judgment is knowing which answer is right.

I didn't learn that at a tech conference. I learned it in a container yard.

The call the sheet couldn't make

Here's a small, ordinary decision I made recently — the kind that never makes it into a case study, which is exactly why it's worth talking about.

A customer needed a pickup. The system had them assigned to a depot that was low on ground stock and dealing with an access issue. The system's logic is clean and narrow: customer is assigned here, stock is low, access is questionable — so send an email, ask the depot for an update, wait on the repair, and keep the move inside the lane you were given.

That's a correct answer. It's also the wrong one.

What I did instead was re-route the customer to a nearby yard I knew was sitting on high stock, with units already staged and ready for pickup. Not a guess — high availability there meant far fewer things that could go sideways on the actual day. Fewer back-actors in play. A clean pickup instead of a hopeful one.

But here's the part the system would never do: I still sent the email to push the original depot on its repairs and access anyway.

Because the customer's pickup wasn't the only thing that mattered. Those idle units back at the first depot still needed to be made ready — for the next order, for the health of the yard, for the delay that hasn't happened yet. So I solved two things at once: the customer's day, and the network's next week.

Why that's the whole point

Hand that exact situation to an automated system and it does one of those things. It re-routes, or it sends the email. It optimizes for the single task in front of it, because that's the task it was given.

The judgment call was seeing that the real answer was "both — in this order, for these reasons." Make the customer happy now. Keep the longer fix moving. Prevent the delay before it's born instead of hoping one never shows up.

That connection — between the job in front of me and the problem three weeks down the line — isn't in any sheet. It's not an answer you look up. It's judgment. And it's the thing we're now trying to teach machines to do, which turns out to be much harder than teaching them to answer.

What I'd take from this

Three things I keep coming back to:

Stop asking whether the AI is right. Start asking whether it decides like you would. A correct answer that ignores next week is still the wrong call. The test isn't accuracy — it's judgment.

The person who can explain their judgment is worth more than the person who just has it. Most of us make good calls on instinct and can't say why. In a world racing to automate, the rare skill is making that silent instinct explicit enough that a system could hold it.

Judgment is mostly about the problem that hasn't happened yet. The sheet tells you what is. Your value is acting on what's about to be — and mitigating it before it arrives, instead of hoping it doesn't.

The part I'm still chewing on

If judgment is the thing we're handing over, then when do you put the guardrails around it? Before you let the system decide for itself — or after you watch it get one call wrong in a place where wrong is cheap?

I don't have a clean answer yet. That's the next piece.

— Real Talk with Riggles