For years, we thought the key brick in autonomous mobility would be the AV stack. It required years of development and massive investment, largely based on rule-based AI.
Then deep learning changed the game.
Today, a growing number of players, I counted 12 as of today, can realistically aim to bring L4 shared autonomous vehicles to the road. The technology itself is becoming more accessible and significantly cheaper — including sensors like LiDAR.
Attention then shifted to the marketplace layer, with customer-facing apps. Uber, for example, positions itself as a layer AV players should plug into.
But this may only be a transition.
A new battle is emerging at the top: devices coupled with AI agents. Big tech companies in the US and China are investing heavily in this space. They are unlikely to replace smartphones immediately, but they will progressively reduce the need to open apps. The agent will select, not the user.
Below this, a critical but often overlooked layer is operations.
Operating fleets of autonomous vehicles requires on top of digital expertise a deep knowledge of local regulations, infrastructure, usage patterns, and strong execution on the ground.
Public Transport Operators have a key role to play here. Their mission should expand beyond collective transport to broader mobility services. In many cases, physics will prevail: low-density areas, off-peak hours, and dispersed demand will continue to require individual or semi-individual transport. The challenge is to combine modes and uses across the full journey.
Another major shift concerns vehicle design.
Just as early cars were “horseless carriages,” today’s robotaxis are in most of the cases still “driverless cars.” This will not last. Vehicles will increasingly be designed for specific services, not as generic cars without drivers.
This means new form factors and new requirements in terms of durability, usage, modularity, adaptibility for upgrades, range, speed, integration in energy and other systems and so on.
Shorter development cycles, customer in the loop and modular architectures will enable true mass customization.
We are moving from privately owned, underutilized cars to fleets of highly utilized vehicles, with long-life platforms and continuous upgrades.
For OEMs, this is a deeper transformation than electrification or software-defined vehicles. It is about designing and managing adaptive service assets.
Ultimately, the real game is systemic.
Each layer — software, hardware, operations, services — cannot be optimized in isolation. Value will come from how well they are combined into a coherent system.
Some players are already spanning multiple layers. Many others are still thinking in silos.
The winners will be those who adapt to this new environment — and there is room, because the scale and diversity of future mobility services will be massive.
