AWAKE / 001 — VOLKSWAGEN. An outside-in application of the Edge Method.
This is not a case study. Volkswagen has not hired me, and I have no access to its internal systems, economics or decision-making. That is the point. A real Awake starts with the CEO’s workflows and data. An outside-in Awake starts with public evidence, separates what is documented from what is only a hypothesis, and commits to a prediction someone can check against reality later. Everything below runs on sources you can click, and it ends with a date on which I will score myself.
It also comes down, in the end, to one paragraph in a Volkswagen press release about engineers coming home to Wolfsburg — and one question: is Volkswagen moving knowledge, or moving work?
The opening question is the method’s first-principles question, applied whole:
If Volkswagen did not exist, would we build Volkswagen?
01 — The trigger
Volkswagen does not have a technology-investment problem. It has CARIAD, cloud infrastructure, AI programmes, a Digital Production Platform built with AWS connecting 43 of its 114 production sites, more than a hundred production and logistics applications, software partnerships on three continents, hundreds of billions in revenue and some of the best automotive engineering talent on earth.
CARIAD, by its own account, generated €1.8 billion of revenue in 2025 — up 34 percent — while producing an operating loss of €2.2 billion, as the Group continued restructuring the unit and replacing technology. Five years in, the software company built inside the old boundaries is growing its deliveries and barely moving its loss.
So the interesting question is not whether Volkswagen understands that software matters. Obviously it does. The question is whether it has built the organizational capability to produce a software-defined car the way we would build one today. Those are very different things. A company can have AWS without being cloud-native. It can have a Chief Data Officer without being able to find its data. It can employ thousands of software engineers without being able to ship software the modern way. It can run hundreds of AI use cases without possessing an AI-native operating model.
Technology assets are not technological capability. That distinction is where this Awake starts — and it is now a rule of the method: never count platforms, licences, teams or models as evidence of anything.
02 — Forget Volkswagen
For thirty minutes, remove Volkswagen from the room. No CARIAD, no MQB, no existing factories, no Audi architecture, no Porsche requirement, no current suppliers, no organization chart.
One mandate: build a global automotive company today. It must design, manufacture, sell and operate millions of safe vehicles. Software exists from day one. AI exists from day one. Over-the-air deployment, modern zonal computing and cloud infrastructure exist. Vehicles can generate operational data continuously; factories can be instrumented continuously; engineering can keep learning from products after they leave the factory.
What would we build? Almost certainly not today’s organizational architecture.
03 — What the clean-sheet company would build
Computational sovereignty
Traditional automotive architecture distributed intelligence across dozens of components and suppliers; each component often arrived with its own controller, firmware and interface, and the OEM integrated it. A software-defined vehicle inverts that relationship: the vehicle has an authoritative computational architecture, components attach to it, and the manufacturer owns enough of the software and interfaces to understand, update and evolve the system as a whole.
Tesla is useful here — not as a model to copy, but as a control group: it says it develops nearly all of its vehicle-control and infotainment software internally and updates its fleet over the air as routine. Rivian is another reference: its second-generation R1 architecture cut the vehicle’s ECUs from seventeen to seven and removed 1.6 miles of wiring per vehicle. But the interesting difference was never the number of computers.
Who owns the architecture?
Who defines the interfaces, changes the software, sees the data, controls deployment, improves a capability after the vehicle has left the factory? That is computational sovereignty, and in the clean-sheet company it is non-negotiable.
Differentiation where the customer pays for it
Volkswagen’s brands are a strength, not a problem. A Porsche should not drive like a Škoda. But first principles force the question: at what layer does differentiation create customer value? Vehicle dynamics, design, materials, brand experience — yes. A different foundational telemetry architecture, a different deployment mechanism, a different identity system per brand? Why? The clean-sheet organization allows enormous variety above a small number of hard architectural boundaries, and below them: one substrate, one deployment model, one telemetry language, one security model, one way for the vehicle to state its condition. Brands differentiate where the customer pays for differentiation — not where organizational history happened to leave a boundary.
The car as part of a loop
The vehicle should not leave the factory and disappear. It enters a loop: it observes reality, produces data, engineering learns, software changes, validation runs, deployment happens, and the loop turns again. The product becomes part of the production system.
Volkswagen saw this years ago. Its 2021 NEW AUTO strategy described a “Big Loop” of millions of connected vehicles feeding real-time data back to continuously improve the fleet, with up to 40 million vehicles on Group software platforms by 2030. That is the right ambition. The Awake question is whether the organization operates that way — and public information cannot answer it. It can only tell us what to test.
04 — Don’t count data. Test observability.
The Digital Production Platform is real infrastructure: 43 factories connected, a renewed five-year AWS partnership, more than a hundred applications. But infrastructure is evidence of spending, not of seeing.
There is a simple test, and it is measured in hours, not months. Pick one production problem from last week. Show me, today: the affected vehicles, the components, the supplier batches, the stations, the software versions, the quality events, the field incidents, and the decisions that followed. Then pick one digital vehicle capability and show the complete loop — field event, data, engineering decision, code, validation, deployment, measured result. Not after a six-month data programme. Now.
If the company can do this, it has something rarer than any platform. If it cannot, the problem is not a lack of data — it is that the organization cannot see itself clearly enough to learn at software speed. Cloud storage does not fix that. Dashboards do not fix that. AI does not fix that; AI exposes it.
Observability comes before intelligence.
05 — A project is not a loop
Volkswagen and Bosch built a serious automated-driving effort: the Automated Driving Alliance put more than a thousand specialists on ADAS software, ran more than 1,500 test and data-collection vehicles across Europe, the US and Japan, and accumulated over 200 petabytes of data. By any measure, that is real engineering at real scale.
And in June 2026, reports emerged that the alliance is being wound down — the technology reportedly judged uncompetitive against the pace of rivals in urban hands-free driving, after roughly €1.5 billion invested, with Volkswagen said to be looking for a partner to source the capability instead. Neither company has officially confirmed it, so hold it loosely. But note what the episode is, structurally: a data-collection project — however large — can be cancelled and replaced by a purchase order.
Compare the operating model, not the technology. Tesla describes learning from field data generated by customer vehicles as an intrinsic property of the product: every deployed unit can contribute to the learning system. A loop that lives in the product cannot be cancelled without cancelling the product.
Is data collection a project, or is learning a property of the product?
The clean-sheet company designs the loop before it designs the organization around it.
06 — Then Volkswagen did something interesting
In 2024, Volkswagen and Rivian created an independent 50/50 joint venture. RV Tech — which by the companies’ accounts employs well over a thousand people — is building a zonal electronic architecture and the software stack for future Rivian and Volkswagen Group vehicles. Volkswagen has said the architecture should underpin its future SSP platform, ultimately supporting up to 30 million vehicles. And in March 2026, production-intent versions completed winter testing in Phoenix and in Arjeplog, Sweden — running in Volkswagen ID.EVERY1, Audi and Scout reference vehicles, exercising all-wheel drive, traction control and over-the-air deployment in extreme conditions.
This matters. Volkswagen has effectively acknowledged that at least one critical capability could be built faster through a new organizational boundary than by continuing as before. That looks remarkably like an edge.
But creating an edge is not enough. The question is what rights it has.
07 — Three rights define a real edge
An edge is not an innovation lab, and it is not a place where interesting people build prototypes for the core. The method gives an edge three rights, granted before it is built.
Architecture rights
The edge owns the design of the system it is building. The core can set genuine constraints and red lines; it cannot keep injecting legacy requirements until the new thing resembles the old thing.
Production rights
The edge performs real work — something the business or customers actually depend on. Not a PowerPoint, not a proof of concept, not innovation theatre.
Inheritance rights
If the edge beats the legacy way against benchmarks agreed in advance, it earns the right to inherit that work. This is the one that decides everything. Without inheritance rights, an edge is only a demonstration.
08 — RV Tech passes two tests. The third is unclear.
From outside, I would score it provisionally like this:
| Right | Public signal | The Awake question |
|---|---|---|
| Architecture | Strong | Can brands or legacy programmes force architectural exceptions? |
| Production | Emerging | Will RV Tech own production-critical capability, or primarily supply technology? |
| Inheritance | Unknown | When RV Tech wins, does the work move into its operating model? |
The first two look promising. The third determines the outcome. And one passage in Volkswagen’s own winter-testing announcement makes me cautious. Volkswagen Passenger Cars is launching a qualification programme in which software specialists train at RV Tech locations, including Palo Alto — and, in the company’s words, “upon returning to Wolfsburg, these specialists will serve as internal experts and bring this expertise back into their development departments as multipliers.” Audi and Porsche are preparing similar programmes.
There is nothing wrong with training. But the sentence reveals an assumption: that the thing that needs to move is knowledge.
What if the thing that needs to move is the work?
If Volkswagen’s problem were mainly that its people don’t understand modern software architecture, sending engineers to the edge and bringing them home would be exactly right. But if the constraints are also supplier commitments, brand-specific requirements, governance, decision rights, legacy platforms, budget structures and organizational interfaces — then the engineer returns, carrying new knowledge, into the system that produced the old result. The core has learned something. The core has not changed.
This is where transformation programmes die, and the sequence is always the same: a new operating model proves itself; the company celebrates; the learnings are documented and transferred; the people return; the old organization adopts what feels compatible and negotiates away everything genuinely disruptive. The edge dissolves. The core survives. The method calls it Edge theatre, and the test for it is one question: if the edge wins, what does it inherit? If the answer is influence, learnings and the right to train the core, the existing operating model remains sovereign.
09 — The map
An Awake produces a map, so here is the outside-in version: the workflow that matters — a vehicle capability travelling from customer need to deployed, learning fleet — with each component placed by how evolved it is, and the antibodies drawn on. (If Wardley maps are new to you: ten-minute primer here.)
Three things on this map are worth more than the rest.
First, the bottom right is the pattern done correctly: cloud infrastructure treated as the commodity it is, bought from a utility, with the production platform built on top of it. No antibody marker there — nobody’s identity is attached to running data centers.
Second, look at the ECU arrow. It points left. Components on a Wardley map drift right over time — custom becomes product becomes commodity — and here is Volkswagen deliberately pulling supplier product back into an owned platform. That is not a mistake; it is the sovereignty move, the same one Rivian made when seventeen ECUs became seven. But a leftward move is expensive and only pays if the platform then moves right as one architecture across the whole group — which is exactly what the 30-million-vehicle ambition claims. The gold arrow and the leftward arrow are one bet, and they only work together.
Third, the antibodies cluster around the platform — precisely where the component is trying to move from custom toward product. That is where antibodies always concentrate: the move that turns a team into a subscription. It is also why the work sits at RV Tech, outside the reach of the old reflexes. The dashed border of that box is where this story will be decided.
10 — The antibodies, by name
I map twelve antibodies inside companies. From outside, honesty requires three lists: what is documented, what is visible in outline, and what cannot be seen without being in the room. None of what follows is dysfunction, and nobody in Wolfsburg is behaving irrationally — each of these protects today’s business, sincerely. That is exactly what makes them antibodies.
Documented. The strongest force on this map is not one of my twelve; it is structural. Under German co-determination, labour holds half of Volkswagen’s supervisory board, and in December 2024 the company and IG Metall concluded an agreement that took plant closures off the table and extended job security commitments toward 2030 in exchange for capacity reductions. I am not scoring that as good or bad — it is the social contract Volkswagen operates within. But it means inheritance rights at Volkswagen are not a management decision; they are a negotiation with people whose explicit mandate is to keep the work where it is. Any honest prediction about RV Tech has to price that in. Also documented: the spin-in history. CARIAD was the previous attempt to build the software capability inside the Group’s own boundaries, carrying the brands’ requirements from day one — five years, repeated restructurings, and a loss that its improving 2025 results barely moved. And the Bosch alliance, if the reports hold, is close to Pilot Purgatory at fleet scale: a 1,500-vehicle proving effort ended before its capability reached the product.
Visible in outline. “We’re Special,” per brand: “a Porsche must not drive like a Škoda” is true at the experience layer and becomes an antibody the moment it reaches down into telemetry, deployment or identity — the scorecard question about architectural exceptions is this antibody’s test. The Procurement Filter: decades of supplier commitments are woven into every platform, each with a contract that outlives any architecture decision. Demo Theater: the risk that the multipliers programme becomes exactly this — expertise returns, is celebrated, and Monday remains Monday.
Invisible from outside. Whether headcount is status in Wolfsburg’s engineering organization, whose budget absorbs the savings when a workflow shrinks, what utilization metrics do to the incentive to automate — the antibodies that live in incentives cannot be read from press releases. That is the half of the map an inside Awake exists to draw.
11 — What I would baseline
A real Volkswagen Awake would not diagnose the entire company; it would select one workflow — vehicle capability → production fleet — and trace one real capability through it: customer need, specification, architecture, software, hardware dependencies, integration, safety validation, brand decisions, production, OTA deployment, telemetry, engineering feedback, next release.
Then baseline it. How long is the complete loop? How many people touch it, across how many organizational boundaries, with how many approvals and supplier interfaces? Who owns the capability end to end? What percentage of its code can Volkswagen change itself? What percentage of the deployed fleet is observable? How quickly does field behaviour become engineering information — and how quickly can the system act on it?
Those numbers would tell us more than the number of AI programmes Volkswagen runs.
12 — The decision
Based only on public information, I would not shut the joint venture down. I would do the opposite: keep it — and change its success criterion.
RV Tech should not succeed when Volkswagen has absorbed its technology. It should succeed as its operating model progressively inherits the work. Give it a real end-to-end vehicle programme, not merely the electrical architecture. Fix the benchmarks against the legacy model before the work starts: development cycle, engineering effort, handoffs, release frequency, defect and rollback rates, supplier dependencies, time from field event to engineering response, fleet observability, cost. Let the existing process run. Let the edge run. Then route work: one percent. If it wins, ten. If it keeps winning, everything. If it loses, kill it. No transformation programme, no declaration of victory — the numbers decide.
And when the edge wins, resist the reflex that feels most responsible: transfer the learnings. Training thousands of people, adoption programmes, aligned brands, updated governance — that is the path where everything disruptive is negotiated away one exception at a time. The alternative is to transfer the next programme instead. Then the next. Resources, people, budget and decision rights follow the work, and the old workflow receives progressively less of it until there is nothing left for it to do.
That is not transformation. That is replacement.
13 — The prediction
My provisional diagnosis: Volkswagen has created a genuine architectural edge. It has not yet demonstrated publicly that it has created an organizational edge with inheritance rights. That distinction will determine whether RV Tech becomes the foundation of a different Volkswagen or simply its most capable supplier. The technology can work and the transformation can still fail.
The decisive moment will not be a software demo. It will come when the RV Tech architecture collides with a brand requirement, a supplier contract, a legacy programme, a governance rule — or the social contract itself.
Watch what happens at the first collision.
If RV Tech is progressively modified to fit Volkswagen, the core is absorbing the edge. If Volkswagen is progressively required to fit the architecture and operating model RV Tech has proven, the edge is beginning to inherit the company. And through it all, one number matters more than any announcement: not people trained, not workshops run, not knowledge transferred — how much real work moves.
The Awake decision
PROVISIONAL: EDGE. The software-defined vehicle architecture should stay protected from the legacy operating model long enough to prove itself on real production work — with one clause added to the mandate: if the edge beats the agreed baseline, it inherits the work. Agree that before the experiment. Afterwards, every antibody is awake, and the hardest negotiation begins only once the edge has demonstrated that parts of the existing organization are no longer necessary.
The number to watch: the share of real vehicle-programme work routed to RV Tech’s operating model.
Revisit: September 2027. I will score the three rights again, in public, against this piece.
What is documented, and what is assumed
Documented, with sources below: CARIAD’s 2025 revenue and operating loss; the Digital Production Platform’s 43 connected factories and the renewed AWS partnership; the NEW AUTO “Big Loop” ambition and the 40-million-vehicle software target; the Automated Driving Alliance’s 1,500-plus vehicles and 200-plus petabytes; the December 2024 agreement with IG Metall; Rivian’s second-generation architecture figures; the completed RV Tech winter testing, its reference vehicles, and the “multipliers” programme, quoted verbatim.
Assumed or hypothesis: whether Volkswagen can run the observability trace today (untestable from outside — that is the point of the test); the internal activity of any antibody; whether inheritance rights exist inside the joint-venture agreements (absence of public evidence is not evidence of absence); the reported wind-down of the Bosch alliance, which neither company has confirmed; and RV Tech’s headcount and the 30-million-vehicle SSP figure, which are the companies’ own statements taken at face value.
What an actual Awake would need
This outside-in analysis can identify the decision. It cannot make it. Inside Volkswagen, I would want one workflow, its economics, its cycle time, its systems, its decision rights, its data lineage and its active antibodies — then we would design it again as if Volkswagen were created today, and let the two versions compete.
Don’t transform your company. Build the one that replaces it.
A note on the method: writing this episode changed it. The observability test of section 04, the three edge rights, and the Edge-theatre warning are now part of the Edge Method. Outside-in episodes are supposed to do that — each one that survives contact with a real company sharpens the instrument.
Frequently asked questions
Is the Volkswagen–Rivian joint venture an edge or an innovation lab?
On public evidence it is a genuine architectural edge: it owns the design of the zonal software-defined vehicle architecture and completed production-intent winter testing in Volkswagen, Audi and Scout reference vehicles in March 2026. Whether it is an organizational edge depends on inheritance rights — an agreed rule that work, not just knowledge, moves to its operating model when it wins. No public document shows that rule exists, and Volkswagen’s “multipliers” programme suggests the current model moves knowledge back to Wolfsburg instead.
What is an outside-in Awake?
An application of the Edge Method to a company using only public evidence: separate what is documented from what is assumed, identify the core-or-edge decision the company faces, and commit to a falsifiable prediction with a revisit date. A real Awake runs inside the company, on one workflow’s actual economics, cycle time and decision rights.
What would show that RV Tech is winning?
Work moving. A complete vehicle programme owned end to end by RV Tech’s operating model, measured against the legacy process on benchmarks agreed in advance — development cycle, handoffs, release frequency, observability, cost — and then the next programme following it. Training programmes, returning experts and steering committees are not evidence of work moving.
Sources
- CARIAD reports strong progress in fiscal year 2025 — revenue €1.8bn (+34%), operating result −€2.2bn.
- Volkswagen Group: RV Tech successfully completes winter testing — March 2026; ID.EVERY1, Audi and Scout reference vehicles; the “multipliers” programme, quoted verbatim.
- electrive: VW and Rivian conclude winter testing programme
- Volkswagen Group collaborates with AWS to transform production for the age of AI — the Digital Production Platform and the renewed partnership.
- How VW Group and AWS are scaling AI across 43 global factories
- Volkswagen Group NEW AUTO strategy, 2021 — the “Big Loop” and the 40-million-vehicle software ambition.
- CARIAD: advanced technology for driver assistance and automated driving — more than 1,500 test vehicles, over 200 petabytes.
- electrive: VW reportedly ends automated driving partnership with Bosch — June 2026, unconfirmed by either company.
- Volkswagen Newsroom: agreement reached — Volkswagen AG positions itself competitively for the future — the December 2024 agreement with IG Metall.
- InsideEVs: the 2025 Rivian R1T and R1S cut 1.6 miles of internal wiring and how Rivian developed its zonal architecture — seventeen ECUs to seven.