Barriers to innovation in German companies come from excessive process control, a lack of budget authority and political reluctance toward new technology, all of which block decisions. Gunter Dueck sees two concrete ways out: targeted industrial policy that builds up entire sectors instead of funding individual projects, and small teams of excellent people who get real freedom to act.
Key Takeaways
- Over-processed companies lose the ability to act: when every trip needs its own approval, management time disappears and the deep concentration that software development depends on is gone.
- AI doesn’t pay off without a foundation: without a unified, machine-readable data foundation, companies just add another expensive layer on top of data silos nobody has cleaned up in 30 years.
- Gifford Pinchot’s rule “Work underground as long as you can” is still the only practical way to push innovation through in a heavily process-driven corporation.
- Industrial policy and subsidies are two different things: subsidies hand out money without direction, while industrial policy names a sector and builds up a whole supplier infrastructure in a coordinated way.
- Humanoid robots are held back mainly by the hand: the tactile sensing needed for different grip strengths is still unsolved, even though the head and movement are technically largely done.
Barriers to Innovation: Why Over-Processed Corporations Grind to a Halt
The biggest barriers to innovation in large companies are homemade. Many corporations are so over-processed that they can barely act anymore. According to Gunter Dueck, that isn’t down to individual people acting in bad faith. It has become a property of the system. What once counted as modern management now paralyzes the organization.
The contrast with the past is stark. Dueck describes how he once hired a renowned scientist at IBM in a single day: boss informed, works council informed, contract signed, done. Today that would be unthinkable.
Budgets worked differently, too. There used to be an annual budget per person, and you had authority over it. Travel, software, social security contributions: everything ran through it without every single expense needing approval. Today every trip has to be approved on its own, and every small amount gets debated.
The damage shows up as lost time. Every fragmented task kicks off a new process, and that eats management time and employee time alike. Nobody thinks twice about pushing things back. Week after week goes by with follow-up questions, objections and another round of alignment.
Deep Focus Is Underrated in Organizations
Deep focus while programming collapses with a single interruption, and getting back into it takes time. Something as simple as a call to lunch can cost a developer around twenty minutes before they are back in the problem.
Dueck compares it to a world chess champion who is asked, in the middle of deep thought, whether he wants meat or fish for lunch. Many organizations still don’t understand this trait of technical people. Instead, highly focused people get ground down by a constant stream of small tasks.
Why AI Often Doesn’t Pay Off for Companies
AI has no standalone business case that pays for itself. The reason is the groundwork it needs: AI needs a foundation. A language model like Anthropic’s has to be fed with knowledge, and on top of that, all company data has to be in a format the model can read and understand.
That groundwork costs a lot of money and brings no direct return at first. Data silos have to go before an AI can make sensible use of company knowledge. That isn’t a new demand. People have been making it for thirty or forty years, and for thirty years it has been politely acknowledged without anything happening.
The bigger risk is that companies simply put a new AI layer on top instead of cleaning up. Then new complicated processes appear, and the whole effort doesn’t pay off. Bolting AI onto messy data structures repeats an old mistake at a higher price.
Germany Has Stopped Measuring Itself Against the Competition
Nobody is embarrassed about coming second anymore, and that is exactly the problem. It used to be a real blow when another country pulled ahead. When Japan overtook Germany in car manufacturing and a Corolla showed less damage than a Golf, that hurt, and it set off a race to catch up.
Today Germans hear at every turn that Scandinavia, the Netherlands or Singapore do better, and they draw no drive from it. Every country used as a comparison comes with a reason to dismiss it: Singapore is a dictatorship, and you’re not allowed to say anything about China. With Finland or Sweden, all that’s left is defensiveness.
Yet the good examples have been there for a long time. Education, unity, determination, energy: other countries demonstrably do these better. The reflex to play down every comparison keeps Germany from learning from them.
Subsidies Are Not the Same as Industrial Policy
The government confuses subsidies with industrial policy, and that is an expensive mistake. With subsidies, billions get released without anyone being clear what they are for. Then programs are put out to tender, a year goes by defining them, and in the end a company takes the grant because it fits the funding criteria.
Nobody asks whether the funded project is any good. Nobody wants to see a business case. Whoever falls into the target industry gets their share, often 20 to 25 percent. The newspapers then report that every euro of funding has triggered a multiple in innovation capital, even though the money was simply a windfall for the companies that took it.
Industrial policy, by contrast, knows what the money is for. It doesn’t fund one company but the growth of an entire infrastructure. It names a sector and creates the conditions for it to emerge.
| Subsidies | Industrial Policy | |
|---|---|---|
| Goal | unclear, broadly advertised | named sector |
| Review | fits the criteria, no business case | targeted build-up |
| Reach | individual companies | entire supply chain |
| Role of the state | hand out money | create conditions |
Special Zones Could Make Innovation Possible in Germany
Special zones are a concrete tool of industrial policy, and Germany barely uses them. China works with innovation zones like Shenzhen, where decisions on flight permits simply get made. Around a thousand flight routes up to 120 meters in altitude have been defined and approved there, so air taxis can be tested and operated between cities of millions.
Dueck applies the idea to German towns. In Bad Füssing, the healing springs are about two kilometers outside town, while the guests stay in town. A closed-off special zone with park-and-ride on the outside and self-driving cars on the inside could connect the two.
The model scales. If it works in one place, it can be applied to tourist regions where you no longer have to stay right on the beach because a car picks you up at your guesthouse. Two municipalities had volunteered for a project like this. The interest is there. What’s missing is the political decision.
Dueck considers the argument that such technology only serves the rich to be inconsistent. Anyone who rejects air taxis because only wealthy people use them would have to say the same about a Porsche or about Mercedes models built for China. As an exporting nation, Germany has to think about its customers, not only about its own use.
Why the Hand Is Where the Business Is in Robotics
The humanoid robot is technically almost finished. The bottleneck is the hand. The head and movement count as largely solved, but gripping is still hard. A robot doesn’t simply know how firmly to hold a bottle without crushing it or dropping it.
Humans do this without effort. They recognize glass or plastic, adjust the pressure and correct it through thousands of tactile signals. Getting that sensing into a robot arm is the real task, and it is big enough that a large corporation devoted only to the hand would be worth it.
Dueck sees an open opportunity for Germany right here. It needs neither its own AI model nor a complete robot, just mechanical engineering and tinkering, which is what German industry considers its strength. Why no German corporation has specialized in this remains an open question.
China Wins Through Coordinated Infrastructure
China’s lead in robotics doesn’t come from individual pieces of software but from building an entire supply chain in a coordinated way. A robot consists of many parts: sensors, actuators, drives. That requires a supplier industry comparable to the automotive sector.
According to this assessment, American companies focus more on pure software, where money flows quickly. China builds up the whole infrastructure together and now develops its own chips meant to replace established suppliers. These chips aren’t an announcement. They are already in current high-end smartphones.
The pace is the real signal. Within one to two years, China will have its own chips, manufactured in Taiwan, and with them independence from Western suppliers. Companies like Xiaomi build smartphones, home appliances, TVs and cars, and are now putting a shared infrastructure underneath all of it.
Intrapreneurship: Work Underground as Long as You Can
If you want to change something inside a corporation, you have to push it through, against resistance if necessary. Dueck points to the principles from Gifford Pinchot’s book on intrapreneuring. The most important ones:
- Come to work each day willing to be fired. Treat each day as if your job were on the line.
- Work underground as long as you can. Stay under the radar for as long as possible before the formal processes catch up with you.
- Work only with the best people. Hire excellent people and stay away from the rest.
The third point is the one most often forgotten today. Instead of pulling together thousands of average employees, Dueck recommends starting with ten or twenty outstanding people, hiring them away from top companies, paying them well and then leaving them alone to work.
“OpenAI did that. Microsoft hired the people and told them they could burn money however they liked. They didn’t even get a budget.”
(Gunter Dueck)
Every corporation knows the opposite model. Innovation sits at the very bottom, several bosses are stacked on top, and as soon as one of them changes jobs, the new one has everything recalculated and stops the funding. That is the opposite of a clear “go."
"Do Nothing”: The Customer Who Doesn’t Want to Change
Many optimization projects fail not because of the technology but because of the customer who doesn’t want to change anything. Dueck talks about his time in factory optimization: despite demonstrable savings of around 20 percent, the big orders never came. A venture capitalist gave him the diagnosis: the main customer simply said “do nothing.”
The same pattern repeats with AI. Real optimization takes bigger leaps: feeding factory lead times into the system, changing how the system works, accepting a higher-level command that reorganizes the processes. Very few are willing to clean up that much.
This resistance isn’t new, but today it is systematized. In the past you could still sort things out case by case. Today the over-processed machinery sides with “do nothing” and makes change even harder.
AI Code No Human Has Ever Looked At
Current AI models generate code of a quality that makes it usable right away. Developers in Dueck’s circle report that Anthropic’s new model is so good that the generated code can be used directly, without a human ever looking at it.
Whole teams of agents now work together on this. You model the roles in a bank, such as loan officers and tellers, replace the people one to one with agents and let them solve a task together. Setups like this have surfaced security vulnerabilities of a kind that hadn’t been seen in purely human work for a long time.
Dueck doesn’t share the fear of incomprehensible machine code. His impression from his own experiments: the code is cleanly documented and explained step by step, often more thoroughly than people would bother to do.
Humanoids Learn Faster Because They Are Built Like Us
A humanoid robot learns faster than a specialized machine because you can train it on human data and human workflows. Once you have built a robot that moves the way we do, ideally all it takes is asking it to watch once and then do the task itself.
That doesn’t work for specialized machines. Every task, such as harvesting asparagus, would have to be designed and programmed specifically. With a humanoid, you can draw on the data that already exists about how people move.
Concrete approaches already exist. At a company near Metzingen, employees wear sensor suits and practice movements on the assembly line, which are recorded and handed to the robot as a skill. The vision goes as far as an app store for humanoid robots, where you rent a skill like asparagus harvesting for a few weeks.
Automation Shifts the Fiscal Balance
When well-paid work is replaced by robots, the balance shifts between the people who finance the community and the people it supports. Dueck uses the term net fiscal contribution for this. Anyone who pays more in taxes than they receive from the state is fiscally positive. Anyone paid mainly by the state is negative in a purely fiscal sense.
The calculation is deliberately provocative and says nothing about the social value of a job. A teacher or a caregiver fulfills a purpose but is negative in the narrow fiscal sense, because the salary comes out of the state budget. The state stays viable only if the positive contributors pay in more than it spends.
The trend points the wrong way. While well-paid jobs become replaceable by humanoids, the state itself pays little attention to AI. Automating well-paid jobs while growing the fiscally negative side at the same time undermines your own financing.
Frequently Asked Questions
What does a company need to prepare before AI can actually be of any use?
AI requires a data foundation: all company data in a format that a language model can read and understand. This preliminary work costs a lot of money and doesn’t yield any direct returns at first, which is why it’s often skipped. Those who instead layer AI over existing data silos only create new, complicated processes. The need to break down data silos has been recognized for thirty to forty years.
How can you tell if a corporation is being stifled by its own processes?
A clear sign is that every single business trip must be approved and even small amounts are subject to debate. In the past, there was a per-person, per-year budget through which travel expenses, software, and social security contributions were processed without individual approval. Every fragmented task triggers a new process and consumes both management and employee time. This isn’t due to the malice of individuals, but rather a characteristic of the system.
How much does a brief interruption disrupt focused development work?
A single interruption shatters deep concentration. A call to lunch can cost a developer about twenty minutes to get back on track. The apt comparison is a chess world champion who, in the middle of deep thought, is asked whether he’d like to eat meat or fish. Many organizations don’t understand this and wear down highly focused people with a barrage of minor tasks.
Why don’t billions in government funding automatically lead to innovation?
Because funding distributes money without direction. First, billions are allocated; then a year is spent defining programs; and in the end, a company walks away with a 20 to 25 percent grant simply because it fits the funding criteria. No one demands a business case or asks whether the project is sound. Industrial policy, on the other hand, identifies a sector and boosts the entire supply chain.
What can special zones do for new mobility technology?
They create a designated area where permits are simply granted. In Shenzhen, around a thousand flight routes up to an altitude of 120 meters have been defined and approved, allowing air taxis to be tested and operated between cities with populations in the millions. The portability of this concept could be demonstrated, for example, in Bad Füssing, where the healing springs are located two kilometers outside the town: park-and-ride on the outskirts, self-driving cars within the town. Two municipalities had volunteered to participate.
Why aren’t humanoid robots in widespread use yet?
The bottleneck lies in the hand. The head and movement are technically considered largely complete, but a robot does not automatically know how firmly it must grip a bottle to hold it without crushing or dropping it. Humans recognize glass or plastic and make adjustments based on thousands of tactile sensations. Integrating this sensor technology into an arm would be a project worthy of a major corporation in its own right.
How can something new still be built within a highly process-driven corporation?
Gifford Pinchot’s principles of intrapreneurship point the way: act every day as if your own job were on the line, work behind the scenes for as long as possible, and work only with the best people. The third point is most often forgotten. Instead of thousands of mediocre employees, ten or twenty outstanding people are enough, well paid and then left to their own devices.
What are the fiscal consequences when well-paid jobs are automated?
The balance between taxpayers and recipients shifts. “Net Fiscal Contribution” means: Anyone who pays more in taxes than they receive from the government is fiscally positive, while those who are primarily paid from the state budget (such as teachers or caregivers) are negative in the strict sense of taxation. This says nothing about the societal value of a job. If well-paid positions in particular are replaced, the state undermines its own financing.
