When I talk to manufacturing leaders about AI, I hear a lot of excitement. I also hear a familiar question: how do we know we can trust it? To me, that’s the wrong question. The better question is, how will you know when you can’t?
I’ve watched this happen before.
Twenty years ago, organizations were asking whether cybersecurity was worth the investment. Security was viewed as overhead. Something you did because regulators or customers expected it. Then the breaches started making headlines, acquisitions started getting repriced, and customers began asking security questions before signing contracts.
Almost overnight, security stopped being an IT problem. It became a business issue. I think AI is heading down exactly the same path. The technology is different, but the pattern is remarkably familiar.
Today, manufacturers are using AI to forecast demand, optimize production schedules, inspect quality, recommend maintenance, improve inventory management and help supply chain teams make better decisions. That’s only the beginning. Before long, AI won’t just recommend actions. It will carry many of them out.
That’s where things get interesting.
For the past two years, most of the conversation has focused on what AI is capable of doing. I think the more important conversation is about something much simpler.
Can you explain what your AI actually did? Not what it was designed to do. Not what your policy says it should do. What it actually did yesterday. Or last month. Or six months ago when a customer asks why a shipment was delayed or why a production decision was made.
That’s the conversation I believe manufacturing is about to have.
One place you can already see it emerging is in due diligence. If you want to understand what businesses truly value, don’t pay attention to keynote presentations or annual reports. Look at what buyers, insurers and enterprise customers ask to see before they’ll move forward.
Those requests tell you exactly where business risk has shifted.
Cybersecurity became valuable because organizations had to prove they were managing it. Today, no serious enterprise customer accepts “trust us” as an answer. They expect evidence. AI will be no different.
We’re already seeing organizations ask about AI inventories, governance practices and accountability. Standards are beginning to emerge. Regulators are paying attention. Customers are starting to ask tougher questions.
None of that surprises me. Manufacturing has been here before.
Manufacturers have never relied on inspection alone to produce quality. They build quality into the process itself. They monitor production while it’s happening. They create traceability because they know finding problems early is dramatically less expensive than discovering them after products have left the factory.
That’s exactly how I think manufacturers should approach AI.
Today, much of the AI governance discussion revolves around policies, committees and compliance frameworks. Those all have a role to play. But if I’m running a manufacturing business, they aren’t the questions keeping me awake at night. It’s:
- Which AI systems are influencing operational decisions?
- What data can they access?
- Which business actions are they allowed to trigger?
- Can I reconstruct those decisions if a customer, regulator or board member asked me to?
Those aren’t governance questions. They’re operational questions. Think about where AI is already showing up across manufacturing.
- A maintenance assistant recommends taking a production line offline.
- A quality inspection system determines whether a finished product is shipped or reworked.
- A supply chain model recommends switching suppliers because of changing market conditions.
- An AI scheduling system reshuffles production across multiple plants.
Today, a person may approve those recommendations. Tomorrow, many won’t.
That’s why I believe manufacturers should stop thinking about governing AI models and start thinking about governing AI actions. There’s an important difference.
Manufacturers have spent decades creating visibility across their operations. They know which machine produced which part. They know what materials were used. They know who approved engineering changes and when they happened.
That level of visibility doesn’t exist because manufacturers expect something to go wrong every day. It exists because controlled processes consistently produce better outcomes. AI should be treated exactly the same way.
Organizations should know where AI is operating, what it’s connected to, what decisions it’s influencing and who remains accountable when those decisions matter. The organizations that build those capabilities now won’t just reduce risk. They’ll move faster. Customers will have greater confidence in them. Procurement conversations will become easier. And leadership teams will be more comfortable expanding AI into more valuable parts of the business because they’ll understand where the guardrails are.
That’s exactly what happened with cybersecurity. At first, governance slowed people down. Then it became the thing that allowed organizations to move faster because everyone trusted the controls.
I think AI is following the same path. Manufacturing has always understood that quality isn’t something you inspect into a finished product. It’s something you build into the process from the very beginning. AI shouldn’t be any different.
The companies that figure this out first won’t win because they have the most AI. They’ll win because they’re the organizations people trust to use it responsibly.