It’s Not the Engineers

Porsche was the most profitable car brand in the world, the cathedral of German engineering. In 2025, Porsche booked roughly €3.9 billion in special charges to unwind an electric strategy it had committed to, with great confidence, only a few years earlier. Automotive operating profit collapsed by 98 percent. The stock lost a third of its value. The company was ejected from the DAX, the index of Germany’s industrial champions.

Seven years earlier, in 2018, Porsche had committed over €6 billion to go electric. Then it reversed course. Add it up and you get something close to a €10 billion round trip that ended roughly where it started.

China was Porsche’s most important growth market for a decade. But while the above unfolded, a Chinese phone company ate Porsche’s lunch. Xiaomi, which had never built a car, launched the SU7: a sedan that looks like a Taycan, outpowers it, and sells at a fraction of the price. In 2024 Xiaomi sold more than 100,000 of them. Porsche sold around 21,000 Taycans. Porsche’s China sales fell by more than a quarter.

That is the wound. Not a margin miss. A hundred-year-old symbol of engineering supremacy, out-iterated by a consumer-electronics company that did not exist in the car business 24 months earlier.

Here is the question that should keep German boardrooms awake at night: does anyone seriously believe Xiaomi’s engineers are better than Porsche’s?

They are not. That is the whole point.

The Comfortable Explanations

When a German industrialist explains the China problem, you hear the same list. Cheap labor. State subsidies. Cheap energy, now gone. A captured home market. Lax safety standards. They copy us.

Some of this is true. Philipp Raasch at Autopreneur makes the sharpest version of the systemic case: China competes as a coordinated, state-directed collective with a decades-long plan, while German firms compete as individual players. That asymmetry is real.

But the comfortable explanations all share a hidden assumption: that if the playing field were level, German engineering would still win on the merits. That the problem is around the engineering; never in it.

I want to challenge that assumption. Not by attacking the engineers, but by defending them.

The Thesis

German and European engineers are among the best in the world. This is not flattery; it is the observable record. They build aircraft, machine tools, fire-suppression systems, and powertrains that run for decades under conditions most startups have never faced.

They are not losing because they are bad, but they are losing because the systems they work inside learn too slowly.

Learning speed is the hidden variable. It rarely appears on a balance sheet, because it is not a cost or a headcount. It is a rate: how fast an organization converts uncertainty into knowledge — how quickly it can try something, watch what reality does, and adjust. A company that learns twice as fast does not work twice as hard. It compounds. Every cycle, the gap widens.

BYD develops a new car in about 18 months. Volkswagen, after a heralded efficiency push, got its cycle down from 54 months to 40. Forty months was announced as a milestone. It is a milestone — in a race being run at 18.

The gap is not effort. German engineers work as hard as anyone alive. The gap is learning rate.

The Proof That It’s Not the Engineers

If the engineers were the problem, you could not change the outcome without changing the engineers. But you can, and the cleanest proof comes from holding talent constant and changing only the system.

Same country. Same labor market. 410× the cost.

NASA needed a heavy-lift rocket, and it funded two. The Space Launch System, built by a traditional aerospace consortium on a specification-first model. And SpaceX’s Starship, built by flying prototypes, intentionally blowing several of them up, and learning from each flight.

By 2025, Starship had flown ten test flights. SLS had flown one. Estimated cost per launch, by the best public figures: roughly $10 million for Starship against about $4.1 billion for SLS: a factor of around 410. Total development cost to that point: about $7 billion versus $23 billion.

Same nation. Same engineering schools. The same suppliers, often literally. The variable that differs is not talent. It is how fast the system is allowed to learn. SpaceX’s own Raptor engine tells the same story at the component level: across three generations it cut engine mass in half and raised thrust, until SpaceX could build on the order of 280 Raptors — for the price of a single legacy RS-25 engine!

Now bring it home to Europe, because the SpaceX story is easy to wave away as American risk appetite.

A European company built a fighter jet this way. Saab Aeronautics developed the Gripen E with somewhere between 2,000 and 4,000 people across more than 100 Scrum teams. A blocker raised by an engineer at the 7:30 morning stand-up reaches the Executive Action Team by 8:30. Decisions that elsewhere take weeks are made in an hour. The aircraft ships new software every six months, at a lower unit cost than its peers. This is a Swedish defense program, about as far from “move fast and break things” as engineering gets, running a fast learning loop on purpose.

A German company did it in safety certification, the one place everyone insists you cannot go fast. Wagner builds fire-protection systems, where any change can invalidate a safety certificate. Instead of treating certification as a review phase bolted onto the end, Wagner built it into the architecture: a model-based modular design with end-to-end traceability, certified code generation, and regression tests run in a TÜV-supervised environment. Certification stopped being a gate you wait in front of and became a property the system carries continuously. Certification time collapsed.

German engineers. German rigor. TÜV in the loop. Faster anyway. The constraint was never the engineer. It was the system the engineer was placed inside.

What “Learns Too Slowly” Actually Means

Slow learning is not laziness. It is a structural property, and it has specific, diagnosable causes.

Late learning is expensive learning. The cost of changing a design does not rise gently as a project moves forward; it escalates. A decision that is cheap to revisit in week three can be ruinous to revisit after the tooling is cut. One anecdote from automotive: a team shaved the flash memory on a control unit to fit the software exactly, saving over $1 million across the production run. After launch, a valuable new feature could have shipped as a pure software update — except it no longer fit in memory. The choice was a hardware recall or killing the feature. Estimated cost of that million-dollar saving: about $10 million in forgone value.

Saving one million dollars cost ten million dollars.

Complexity has outrun the old ways of learning. A colleague in automotive (who, understandably, does not want to be named) spent months chasing a fault where a high-end car’s powered trunk would reopen a few seconds after closing, but only when luggage sat in one particular spot. The signal propagated across so many networked controllers that the team could not trace it in any reasonable time. A (traditional) modern car carries 80-plus controllers and tens of millions of lines of code. You cannot reason your way through that on a whiteboard. You have to learn your way through it: fast, with integration and feedback built into how you work.

This is why 40 months is not just a slower version of 18 months. They are different machines. One is built to specify, then build, then find out. The other is built to find out continuously.

Why the System Learns Slowly — and Why It Isn’t the People’s Fault

The German engineering tradition is the product of a world that rewarded it. For decades, the winning move in safety-critical physical products was to specify completely, plan rigorously, gate every transition with a formal review, and lock the design before building anything. The V-model. It produced reliable, certifiable, world-beating results. But only for problems that were well understood before development began.

That tradition carries a built-in assumption: the problem is known up front. When that holds, rigor is a superpower. When it does not, the same rigor becomes a brake. The cost of change discourages the very experiments that would generate the missing knowledge. Compliance hardens from evidence that the product is good into a gate you must pass before you are even allowed to find out. Hardware decisions lock in early and freeze the software around them.

The V-Model expects the problem to be known up front.

None of this is an engineer being slow. It is a system, optimized for a previous era, doing exactly what it was built to do — in a race that changed underneath it.

This is also why Raasch’s “individual players versus a coached collective” cuts deeper than it first appears. China did not merely coordinate companies. It built an industrial system tuned for fast iteration: common platforms, regulation that made over-the-air updates accountable instead of forbidden, and engineers who carried a smartphone-iteration reflex straight into cars. The Germans are being out-learned, not out-thought.

What Fast-Learning Systems Actually Do

The cure is not “try harder” or “hire better.” The engineers are already excellent. The cure is to rebuild the system around learning speed. The same pattern shows up in every example above:

• Validate demand before you scale. GE’s Opal nugget-ice maker was built in months for around $50,000 and sold roughly 6,000 units in 30 days through crowdfunding — customers paying upfront for a product that did not yet exist. A comparable earlier appliance had taken three years and a large investment to sell 20,000 units in a year. Same company. Different learning loop.
• Integrate continuously, and fly. SpaceX learns from real flights, not from a final integration phase that arrives years too late.
• Make assurance a byproduct, not a phase. Wagner generates certification evidence continuously, with the regulator inside the loop rather than waiting at the end of it.
• Make organizational latency illegal. Saab moves a blocker from engineer to executive in an hour, not a quarter.

These moves have names. In my book Product Velocity, they are four principles: Value Thinking (align everyone on the outcome, as GE did), Architect for Flow (modular structure so teams learn in parallel without colliding, as Wagner and Saab did), Shift Left (move learning earlier, where it is cheap), and Accelerate (treat shipping as the start of learning, not the end — the way Rolls-Royce sells engine-hours off live telemetry instead of selling engines and walking away).

None of them require better engineers. All of them require a faster system.

The Good News Hiding in the Wound

If Germany were losing because its engineers were bad, the situation would be close to hopeless. You cannot re-grow a century of engineering culture inside a single product cycle.

But that is not the diagnosis. The talent is here, proven, world-class. The constraint is the operating model around it — and operating models are something you can change deliberately, as Saab and Wagner, and even a reluctant, adapting Volkswagen inside China, are already showing.

So the wrong response to Porsche’s €10 billion round trip is to demand that the engineers work harder and the auditors audit more. That is pressing harder on the brake. The right response is to ask a colder question — the one that actually decides these races.

Not “is it good enough?” German engineering won that argument a long time ago.

But: “can we learn fast enough to get there before the customer stops waiting?”

The engineers were never the problem. The only open question is whether the system around them can learn in time.