Velox Robotics: Structure Follows Culture

The Garage Days

Jonas Richter left the Fraunhofer IPK in Berlin in the spring of 2019 with a prototype in his backpack and a conviction in his head: the world needed a better micro-assembly robot. The machines on the market were either too expensive for small-batch production or too imprecise for the components that semiconductor and medical device manufacturers needed. Jonas had spent eight years at Fraunhofer proving that autonomous robotics could solve this. Now he wanted to prove it could be a business.

He rented 200 square meters in a converted factory building in Berlin-Adlershof, the city’s science and technology park. The first twelve people were researchers, engineers, and one business psychologist named Lena Vogt, whom Jonas hired because a friend told him every startup needs “someone who understands people.” He did not fully understand what that meant. He would learn.

The early months were electric. No titles, no processes, no org chart. If someone had an idea, they built it. If a customer expressed interest, the entire team rallied around the opportunity. Their first real breakthrough came when Nexion Semiconductor, a chip manufacturer in Dresden, needed a robot that could place components smaller than a grain of sand onto circuit boards at production speed. Jonas flew to Dresden, demonstrated a prototype on the factory floor, and came back with a letter of intent. The team celebrated with pizza and stayed until 2 AM improving the calibration algorithm.

The chaos was productive. Ideas collided, merged, and sometimes died within the same week. Technical solutions were passionately developed, tested, and often discarded. Experts were consulted, and if they fit, they joined. It felt like the most natural way to work. Structure followed culture. The informal roles, the improvised processes, the ad hoc decisions: they all grew from the culture Jonas and his team had built. Nobody designed it. It just happened.

Until it stopped working.

The Bet

By the time Velox Robotics had 20 employees, the naturalness had turned into frustration. Three teams were working on overlapping problems without knowing it. A critical sensor integration had been built twice, differently, and neither version worked with the other. Customer requests sat in email inboxes because nobody was sure whose responsibility they were. Jonas spent his days resolving conflicts between engineers who each thought their approach was the right one.

The business needed capital to grow. Jonas pitched to investors and found two who believed in the product. Dr. Stefan Weber, a managing partner at Capital Ventures with a background in deep tech, brought financial discipline and a network of industrial contacts. Anika Berger, an angel investor and former VP Engineering at KUKA Robotics, brought something harder to find: operational experience in scaling a robotics company. Both became co-owners.

With investors came expectations. Growth targets. Quarterly reviews. Jonas felt the pressure shifting from “build something brilliant” to “build something that scales.”

It was Lena who saw what was coming. She had watched the creative chaos turn corrosive. The organic culture that made the early months magical was not surviving the transition to 20 people. She had studied organizational development at Humboldt-Universität Berlin. She knew the research. She also knew Jonas.

“You need to stop writing firmware,” she told him one Tuesday morning, standing in the kitchen while he debugged a sensor driver on his laptop. “You need to lead this company.”

Jonas looked up. “I am leading this company.”

“You are solving technical problems. That is not the same thing.”

What followed was a conversation that lasted three hours and changed the trajectory of Velox Robotics. Lena had been reading about AME3, an enterprise operating model that separated leadership into three functions: the Owner, who drives product success, the System Lead, who builds the work system, and the Team, which delivers customer value. She argued that Velox needed this separation now, before growth made the chaos permanent.

Jonas resisted. He was a researcher. Frameworks felt bureaucratic. Lena did not push. She maneuvered. She suggested they bring in an external facilitator to “just explore the idea.” She framed it as an experiment: one workshop, one day, all 20 people. If it did not work, they would drop it.

The AME3 Setup Workshop happened on a Friday in October. By the end of the day, the company had an Ambition (the business case Jonas had presented to the investors), first Goals, and a clear structure. Jonas was the Owner. Lena was the System Lead. The Teams organized following LeSS, the Large-Scale Scrum framework that Lena had studied. An external System Lead was engaged to facilitate the transition.

Jonas later admitted that Lena had maneuvered him. “She let me think I was still making the decisions. I was. But she had already decided what decisions needed to be made.”

Jonas believed that structure follows culture. Build a great culture, and the right structure will emerge. That had been true at 12 people. It was breaking at 20.

Lena believed the opposite. She had borrowed it from the organizational researcher Craig Larman: “Culture follows structure” (Larman’s Laws). In large groups, the structure you create determines the culture you get. You cannot preserve an organic culture by wishing for it. You have to design the structure that lets it survive. The startup had grown past the point where structure could follow culture. From now on, culture would have to follow structure.

The Growth

Scaling from 20 to 50 employees tested every assumption. The naturalness of the founding years was in danger of being lost. New hires did not share the founding team’s intuitive understanding of how things worked. They needed clarity: who decides what, how priorities are set, where their work fits into the bigger picture.

AME3 with Scrum and LeSS provided that clarity without replacing the culture Jonas valued. Teams were empowered to make decisions independently. All work was self-managed. The structure created space for autonomy rather than constraining it.

Lena made a decision that seemed costly at the time: every employee, new or existing, received leadership and communication training. Not a one-day workshop, but an ongoing investment. She argued that self-managing teams need people who know how to communicate under pressure, give feedback, and resolve conflicts without escalation. The results showed within months. Teams that had been stuck in unproductive arguments began resolving disagreements on their own.

As the company grew, Lena could not do everything herself. Four more System Leads grouped around her, each with a specific focus. Kai Lindstrom, a German-Swedish organizational psychologist, took over recruiting and ensured that hiring decisions were made collectively by Team representatives, the Owner, and the System Lead. Tobias Krause, a former Siemens engineer, focused on engineering practices and technical quality. Priya Sharma, with an MBA from IIM Bangalore and product management experience at Bosch Rexroth, coached Jonas on product strategy and facilitated Wardley Mapping sessions. Marie Dubois, a French-German psychologist and former agile coach at Zalando, focused on team coaching and cross-team coordination.

The investors noticed. During the quarterly Tournaments, Stefan Weber and Anika Berger no longer saw slide presentations about planned features. They saw working product demonstrations. Teams showed what they had built, not what they intended to build. Anika, who had lived through KUKA’s scaling challenges, told Jonas after one Tournament: “This is what I wish we had built at KUKA. Structure that lets people do their best work.”

The Dependency Crisis

At 50 employees, what had been manageable at 30 broke.

The product, which they had named MIRA (Micro-Assembly Intelligent Robot Arm), was growing in complexity. The platform needed to handle different assembly tasks across multiple industries: semiconductor, medical devices, and precision optics. Each team had taken ownership of a specific capability. One team built the vision system. Another developed the motion planning algorithms. A third handled the sensor integration layer they called SensorGrid.

The problem emerged gradually, then all at once. Teams began building isolated solutions to avoid waiting on other teams. The vision team created its own sensor interface because the SensorGrid team was three weeks behind. The motion planning team bypassed the standard API because it was too slow for their use case. Each decision made sense locally. Collectively, they produced a product that could not be integrated.

The breaking point came during a demonstration for a major automotive supplier. The MIRA platform was supposed to show a complete micro-assembly cycle: vision, sensing, motion, and placement in a single integrated flow. Three teams had each built their part. But the parts did not fit together. The SensorGrid module that one team had built was incompatible with the interface the vision team expected. The demo crashed. In front of the customer.

Jonas called an emergency meeting. He was furious. “We have 50 people and we cannot do what 15 people did two years ago.”

Lena was calm. “The problem is not the people,” she said. “It is the architecture. The product architecture reflects the organizational structure. That is Conway’s Law. If we want the product to be integrated, the teams need to work on shared components, not isolated ones.”

This was the hardest conversation Velox Robotics had ever had. A full-day workshop with all System Leads and team representatives. On the table: the entire product architecture. The question was not what to build next. The question was how to restructure the product so that teams could work independently without creating incompatible components.

The workshop produced a redesign plan. Components were reorganized around stable interfaces. Teams were restructured to minimize overlap. The SensorGrid module was split into a shared library maintained by all teams and a set of team-specific adapters. It was painful. Some engineers had to leave work they loved and take on responsibilities they did not want. But the alternative was a product that could not ship.

The redesign took three Matches to implement. Integration failures dropped by 70% in the following quarter.

The AI Shift

The transformation happened quietly. Tobias, the System Lead for engineering practices, introduced AI coding agents for firmware development almost as an experiment. He set up a pair programming session where an engineer worked alongside an AI assistant to write the repetitive sensor initialization code that every new hardware integration required.

What had taken two days of tedious, error-prone work was done in three hours. And the AI-generated code was cleaner, because the agent applied the Team’s coding standards consistently. Tobias showed the results in the next retrospective.

Not everyone was convinced. Two senior engineers pushed back. They argued that AI-generated code was a black box, that it would create dependencies the Teams could not maintain, that it would erode the engineering craft they had spent years building. Lena did not overrule them. She asked Tobias to set up a controlled experiment: one Team used AI agents for a full Match, another did not. After one month, the results were clear. The AI-assisted Team delivered 40% more Improvements with fewer defects. The skeptics did not become enthusiasts overnight, but they stopped blocking. Within two months, every Team was using AI coding agents for routine tasks.

The deeper change was not about speed. It was about scope. With AI handling the repetitive engineering work, teams had capacity for things they had previously handed off. One team that used to deliver only firmware began managing the supplier relationship for its sensor components. They ran their own quality validation. They negotiated delivery schedules directly with the supplier. The boundary between “development” and “operations” dissolved.

Lena observed that the AI tools changed how retrospectives worked. Teams now had data: cycle times, defect patterns, integration frequency, build stability trends. The conversations shifted from “what felt wrong” to “what does the data show, and what should we try next.” The System Leads used this data to facilitate sharper, more focused retrospectives.

During a Tournament, Priya presented an updated Wardley Map of the MIRA product. Several components that had been in the Custom Built zone had moved toward Product. The AI tools had accelerated evolution. Stefan Weber, the investor, recalculated the growth valuation based on the team productivity data. The numbers surprised everyone, including Jonas. The company was producing more value per engineer than comparable startups twice its size.

Anika Berger looked at the numbers and said: “You are not a 50-person company performing like a 50-person company. You are a 50-person company performing like a 100-person company. That is your competitive advantage.”

Structure That Breathes

Velox Robotics at 50 employees looked nothing like Velox Robotics at 12. Seven teams. Five System Leads. One Arena. Quarterly Tournaments with investors who saw working products, not promises. A product architecture designed for independent delivery. AI tools woven into the daily work.

But it felt the same.

New engineers joining the company remarked on the energy. The speed of decision-making. The absence of bureaucratic overhead. The feeling that their work mattered and that someone cared about both the product and the people building it.

That was not an accident. It was the result of a bet that Lena Vogt had made two years earlier, standing in a kitchen, telling a robotics researcher that he needed to stop writing firmware and start leading.

In the garage days, structure had followed culture. Twelve people, no titles, no process. The culture was everything, and it shaped whatever structure existed. Jonas had believed this could last forever.

Lena had known it could not. At scale, the equation reverses. Culture follows structure. Build the right structure, and the culture you want has room to grow. Build the wrong one, and no amount of motivational posters or team-building offsites will save it.

The art is knowing when to flip it.

Jonas still occasionally opened his laptop to debug a sensor driver. Lena let him. Some things should stay organic.