Why do companies hesitate to adopt AI?
Recently, I attended a startup association event in Berlin where participants shared findings on AI use in companies and startups. One figure stuck with me: only 3% of companies use AI in their internal processes.
When a trusted source cites a number, skepticism is healthy. A low figure is often even lower, and a high one higher—unless it is an obvious outlier. So why 3%? Why so low? With all the hype and capital flowing into AI, why are companies—usually eager to save time and money—so hesitant?
The answer lies in the production system: it is not a playground for new technology. It demands reliability, not experiments.
The Nature of a Production Process
At first glance, the hesitation seems puzzling. If AI can handle repetitive tasks and cut costs, why not use it? A closer look reveals the logic of production systems.
Manufacturing is a finely tuned sequence of tasks, a tightly balanced system where supply chains, materials, and tools must interact flawlessly.
Automakers coined the term production hell for this reality: a brilliant car design is meaningless if production stumbles. Once a design is frozen, even a single change can disrupt the entire process. You are forced to make it work as planned.
The system punishes deviations. It is built for stability, not for experiments.
False Assumptions About AI and Production
Solving production problems can take weeks or months.
Consider a legendary story about a critical rocket part:
A supplier quoted $100,000 for a component. The CEO remarked, “This is no different from a garage door actuator—it shouldn’t cost more than $3,000.” He tasked a talented engineer to hit that target.
Nine months later, the mission succeeded. The engineer proudly sent a five‑page report describing every twist and turn of his creative process and how he finally reduced the cost to $4,500. The reply from his boss came five minutes later: “OK.”
No praise. No fanfare. Yet every rocket launch now saved $95,500.
That is production hell.
Could AI have helped? Yes—but only before the manufacturing process is set. Once production is locked in, there is no simple way to optimize that actuator process.
This story illustrates why AI adoption in production is so difficult. It rests on two false assumptions:
A universal, ready‑to‑use AI exists that can seamlessly step into human roles.
Production processes are flexible and can easily accommodate new tools.
Both are wrong, at least in the world we know. Production is a dense web of interdependent compromises. Humans, machines, and processes form a resilient network.
Assuming that an external tool like AI can just “drop in” and improve it ignores this systemic logic.
The system is designed to deliver uniform results at minimal cost, not to host experiments.
The old wisdom applies: “Never change a winning team.”
In production, this translates to: don’t touch a running process unless absolutely necessary.
The Cost of Disruption
Companies cling to their established processes for a reason: a change can be catastrophic.
A system interruption is a wrench in the gears:
Costs spike
Sales stall
Revenue stops
Even one day of downtime can burn millions, cost jobs, and alienate customers.
AI therefore introduces not just technical risks, but existential ones.
Adopting AI isn’t about swapping a tool—it can mean rethinking the product, the tasks, the process, and the entire supply chain.
Destabilize the system, and it retaliates immediately.
The Illusion of Progress
Many observers misinterpret AI as a single, mature tool.
In reality, AI consists of many evolving models, often focused on tasks outside traditional production.
The hype promises breakthroughs, but the system’s logic is simple: any newcomer must outperform the current setup under all constraints.
Until it does, it stays outside the system.
AI evangelists often claim that “data is the oil of the 21st century”, promising:
“Give us your data, and we will eliminate inefficiencies and cut costs.”
But production veterans, hardened in real production hell, remain skeptical.
Humility and experience make them doubt such sweeping claims.
The Role of IT and Pragmatic Skepticism
Inside companies, IT departments—informatics and computer science teams—already manage the data flow that keeps production running smoothly.
So why don’t they jump on the AI train? Are they failing?
No. If the system runs reliably, they are doing their job.
When AI proponents ask, “When will you add AI?”, IT experts respond with hard, practical questions:
How exactly will AI optimize step 12?
What does it do with our data?
What measurable outcome will it produce?
Prove the gain.
As guardians of stability, IT professionals protect the production system from reckless experiments.
The Market as the Driving Force
The smartest move for companies right now is to wait and monitor the market.
Producers who survive in competitive markets already know their craft—or they would have disappeared.
Change comes from outside, driven by the market:
Customers abandon products (think telephone books after the internet, or digital cameras after smartphones).
Technology shifts force adaptation.
Internal change is rarely voluntary.
The system bends only to force. It does not move on enthusiasm alone.
AI will only be adopted when external pressures make it unavoidable.
Until then, IT departments treat it with cautious curiosity.
Conclusion: The Wisdom of Restraint
So, why the hesitation?
Because production is not an experimental lab.
It is a system optimized for stability, and change comes only when forced.
Current processes work as intended—AI does not yet fit.
That will change with new products and the next round of production hell, but not now.
Manufacturers succeed because they master their process, technology, and supply chain.
The low adoption rate of AI reflects pragmatism, not backwardness.