The most useful thing I heard at Applied AI for Distributors wasn't about a model. It was six words from Matt Sigelman of the Burning Glass Institute: "AI is really easy to use."
He's right. And that's the problem.
When something is easy, everyone has it. And when everyone has it, it stops being an edge. So the real question all three days wasn't should we use AI? It was how do we govern it?
The AI is the easy part. The judgment you wrap around it is the hard part — and that's where the advantage lives now.
None of this was new to us. We've been saying it for years: that companies were "testing in production" with their biggest decisions, and that AI's real risk isn't the hack — it's the governance gap.
The conference just put three fresh faces on the same idea.
AI opens the door — and the attack surface
IBM's Jeff Crume gave the cybersecurity keynote. It was a useful cold shower.
A dealership chatbot agreed to sell a car for $1. Nobody hacked it. Someone just typed the right words, and the bot did as it was told.
That's prompt injection — hijacking a system with a sentence. Indirect injection hides the same trick inside an ordinary email or document, ready to fire on its own.
Then there are deepfakes. They're good enough now that companies have wired tens of millions to a voice or a face that was never real.
Crume's point was plain. You can't trust an output, a voice, or a video on sight anymore.
We've argued this for a while. In our notes on AI security, the rule is blunt: security is a posture, not a purchase.
Let the data decide where a model call runs. Public work goes to a public API. Confidential data stays in a private tenant — or a model that never leaves the building.
"Don't trust the output" and "don't trust the wire" are two halves of the same job. The model is cheap. The verification around it isn't.
The exception that quietly became the price
One session stuck with me. It was about pricing.
The idea is simple. A pricing exception five days old is a decision. Ninety-five days old, it's a price.
You approve a 6% discount during a cost spike in March. By June the cost is back down. The discount is still there. And the customer now expects it.
Worse, it's buried — across your ERP, a special pricing agreement, a CRM note, a spreadsheet. No one is tracking what it costs you.
The fix isn't smarter AI. It's a default. Exceptions expire unless someone renews them on purpose.
This is the same lesson from our digital-twin work — Your Best Decision This Quarter Will Be One You Made Twice. Most companies run their most expensive experiments straight on the live business, then read the result off the P&L weeks later.
That's not strategy. That's testing in production. A discount that drifts for 95 days is the same mistake, slowed down.
Same cure both times: rehearse the decision, then govern the default. Make the safe path the automatic one.
AI redefines jobs — it doesn't replace them
Back to Sigelman.
His session argued AI will reshape distribution roles, not erase them. As routine work gets automated, the value moves to judgment — planning, deciding, knowing when the AI is wrong.
"AI is really easy to use," he said. The hard skill is trusting it at the right moment.
This is why the wrap-around matters. When a capability costs almost nothing and everyone has it, it stops being an advantage — what we call the commoditization of expertise.
If you and your competitor run the same models, the model isn't the edge. The people who catch the bad price, the injected instruction, the confident-but-wrong answer are.
Your workforce is the verification layer.
The one thing to take home
Put the three together and they rhyme.
Crume says don't trust the output. The pricing session says don't let a decision drift. Sigelman says the value is the judgment, not the tool.
One idea, three angles. In 2026, distributors stop asking should we use AI? and start asking how do we govern it? The ones who build the judgment layer win.
The AI was never the hard part. The hard part — the moat — is everything you build around it.