Ask a plant manager what they would do with AI, and the answer is rarely "write my emails."
It is: Which orders are going to ship late this week? Where are we short on material? What did our biggest customer change in last night's purchase orders?
Every one of those answers lives in a system that runs the plant: the ERP, the production database, the EDI feed from the big account, the data coming off the floor. And that is exactly where most AI tools stop.
The problem: AI can't reach the systems that run the plant
A year ago, the honest question about AI in operations was whether it was good enough. That question is largely settled. The models have improved dramatically, and AI agents that plan, use tools, and work through multi-step tasks have moved from demos into daily work. The people asking for more are no longer only software teams. They are planners, customer service leads, and operations managers who use AI every day and want it to do more than draft text.
What they run into is a wall. The systems that hold the answers were installed over decades, many of them on servers inside the building. None of them are broken. They are reliable, paid for, and full of exactly the data an AI agent needs. They were simply never built to be reached by AI, and most AI tooling was built for modern cloud applications, not for them.
So ask the AI how much stock you have, and the best it can do is read last week's spreadsheet that someone exported and pasted in. It can't check the ERP for what's on the shelf right now.
Why it's a problem
The value is real, and it's going unclaimed. The pattern was already visible late last year. McKinsey's State of AI survey (November 2025) found that among organizations using AI in manufacturing, 56 percent reported cost decreases, tied for the highest of any business function. Yet manufacturing had the lowest share of regular AI use of the 11 functions McKinsey tracks, 10 percent, and 91 percent of respondents said their organizations weren't using AI agents there at all. The value was proven where AI was connected. Almost nobody had connected it.
The gap is widening. Since then, the models have raced ahead. The connections to plant systems haven't. Every improvement in AI makes the cost of being disconnected larger, because the AI could do more if only it could see what's happening in the plant.
The knowledge lives in people, not systems. In most plants, the integration isn't in the software. It's in a person. They know that the part number on the customer's purchase order and the part number in the ERP are the same part. They know which revision is current and which spreadsheet is the real one. An AI agent can't ask them every time, and when that person retires, the rules leave with them.
Replacing the systems isn't the answer. The usual advice is to modernize first. That takes years, costs a fortune, and puts AI value at the end of a project most plants will never finish.
After 30 years of connecting enterprise software, I'll say it plainly: in manufacturing, the roadblock to AI is no longer the AI. It's integration.
What PipeGears delivers
PipeGears is AI agent infrastructure for manufacturing. It lets the AI agents your team already uses (Claude, ChatGPT, Copilot, and others) answer questions and act on data from the systems that run your plant, wherever those systems live, in the cloud or behind your firewall. Nothing gets replaced. Your IT team decides what the AI can reach.
Here's a typical morning in a plant running PipeGears.
At a cabinet manufacturer, the production planner asks the AI: "Which orders due Friday are missing material?" She gets a short list: four orders, the missing components, and the supplier for each, straight from the plant's own ERP.
A customer service rep asks: "What did our largest customer change in last night's purchase orders?" The answer comes back with the three changed lines and the orders they affect. The customer's part numbers and the plant's part numbers already match, because that knowledge is now written into the system instead of living in one person's head.
Nobody exported a spreadsheet. Nobody waited for a report. The ERP is the same ERP. It just answers questions now.
Why it's valuable
- ✓Faster decisions. The questions that used to take a report request or a morning of spreadsheet work get answered in the moment they're asked.
- ✓Your people on the work that needs them. Your most experienced people stop being the human link between systems and spend their time on the exceptions that actually need judgment.
- ✓Knowledge that stays. What they know gets written down once and carried by the systems, so it doesn't walk out the door at retirement.
- ✓More return on what you already own. The ERP and databases you've paid for become more valuable, not obsolete. No rip-and-replace, no multi-year project before the first result.
- ✓Every AI improvement pays off. Once your AI can reach your systems, each new generation of models makes it more useful to the plant instead of widening the gap.
The bigger point
The next few years of AI value will not go to the companies with the best model. They will go to the companies whose systems their AI can actually reach.
Most of the systems that run manufacturing can't be reached by AI today, through no fault of their own. Making them reachable, without replacing them, is what PipeGears does every day.
That is what we mean when we say: talk to your factory.
Want to see AI answer a question from your own systems? Book a 30-minute walkthrough at pipegears.com/contact.
Source: McKinsey & Company (QuantumBlack), The state of AI in 2025: Agents, innovation, and transformation, November 2025. Global survey of 1,993 participants, June 25 to July 29, 2025. Figures cited: manufacturing function 10% regular AI use (Sidebar exhibit, p. 9), 91% not using agents (Exhibit 2, p. 5), 56% cost decrease (Exhibit 7, p. 12).