The automotive industry is rapidly moving from traditional vehicles to Software-Defined Vehicles (SDVs), with AI emerging as a critical layer of intelligence. In your view, what does the transition from an SDV to an AI-defined vehicle really mean, and how fundamentally could it change the way vehicles are designed, developed and experienced?
An SDV makes the vehicle programmable while and AI-defined vehicle makes it perceptive, adaptive and increasingly capable of informed decision-making. The shift is less about adding features and more about changing where a vehicle’s behaviour comes from. In a traditional or SDV programme, engineers define requirements, decompose them into functions, implement them and verify against the specification. In an AI-defined vehicle, behaviour is shaped by data, objectives and constraints, so the design process is organised around those instead.
This changes how vehicles are designed and developed:
- Design around data and compute. Vehicles are designed around data flows, compute capacity, sensors, AI models and updateability, not only mechanical packaging and fixed feature sets.
- Continuous delivery. Development cadence moves from multi-year platform cycles to continuous releases, with ML engineering and MLOps capabilities sitting alongside classical automotive engineering.
- ML infrastructure as core infrastructure. Data lakes, labelling pipelines, training clusters, experiment tracking and model registries become as central as the test track.
- End-to-end teams. Teams reorganise around capabilities such as perception, planning or cabin intelligence, rather than around individual ECUs.
For the customer, the difference is tangible:
- Genuine personalisation. The car learns routines, preferences and driving style instead of offering a settings menu.
- Agentic interaction. Conversational assistants can act (navigation, climate, scheduling, service booking, explaining a warning light) instead of responding to rigid voice commands.
- Predictive behaviour. Maintenance happens before failure, range and charging plans reflect the owner’s habits, and comfort is tuned to context.
In a nutshell: SDV was about making vehicles updatable; AI-defined is about making them adaptive.
SDV transformation is often viewed primarily as a technology and engineering challenge. From your experience, what changes are required across organisational culture, talent, product development and business strategy for traditional automotive companies to successfully embrace a software-first model?
SDV programmes fail more often on organisation than on technology. Volkswagen’s Cariad unit is the most visible example of how hard it is to build a full stack, a new architecture and a new organisation at once inside a hardware-centred company, while Tesla and several Chinese EV makers have shown how much faster teams can move when software leads the programme. The hard part is not writing more software. It is turning an organisation built around hardware programmes, long development cycles and fixed release milestones into one built for continuous learning and delivery. India’s growing SDV engineering ecosystem shows that the talent base exists, but companies must convert engineering capacity into product ownership, IP and end-to-end accountability.
Culture. Traditional automotive is optimised for getting it right once: “freeze and certify”. Software culture is optimised for getting it better continuously: iterate and learn. Both are valid, but they conflict. Software quality improves when failures are surfaced early and analysed without blame, which is difficult in hierarchical organisations where bad news travels slowly.
Talent. Siloed mechanical, electrical and software skills must give way to cross-functional teams that combine AI, cybersecurity, safety, cloud and product management with systems thinking. Software architects, platform engineers, ML and data engineers, security specialists, product managers and UX researchers are roles that traditional OEMs did not have in volume. Reskilling existing systems engineers, who understand both vehicle physics and software, matters as much as hiring.
Product development. Features are no longer finished at launch, and software can be updated continuously. Instead of redeveloping software for every vehicle programme, one common platform should serve multiple models and brands. Agile must be adapted for regulated, safety-critical systems: continuous integration, automated testing, simulation-based validation, and compliance evidence generated as a by-product of development rather than as a final gate. Co-creation across OEMs, Tier 1s, hyperscalers, startups, semiconductor companies and technology firms is essential to optimising cost and time to market.
Business strategy. The traditional model earns most of its profit at the point of sale and through parts and service. A software-first model adds recurring revenue: subscriptions, feature activation, services and, potentially, data-driven offerings. The evidence so far is mixed. Customers resist paying for features their hardware already supports, and removing or locking features after sale damages trust. Offers that create clear, ongoing value, such as connectivity, navigation, safety and energy services, tend to land better than artificial paywalls.
As connected vehicles continuously generate and process vast amounts of data, how will AI transform the vehicle from a product that is simply operated by the user into an intelligent platform that can continuously learn, adapt and evolve throughout its lifecycle?
AI turns the connected vehicle from a fixed, user-operated product into a continuously learning mobility platform. It senses context, interprets behaviour, predicts needs, adapts locally in real time, and improves fleet-wide through secure cloud learning and over-the-air (OTA) deployment. The key is a closed loop: vehicle data to cloud insight, then a validated model or software update, then back to the vehicle, all within strict safety, privacy and cybersecurity boundaries.
A connected car can generate many gigabytes of data per hour, depending on its sensor suite, so indiscriminate uploading is neither economical nor necessary. Edge filtering and event-triggered data collection become essential.
Learning happens at three levels:
- The individual vehicle adapts to its owner and its own condition. It learns routines, cabin preferences, driving style, typical routes and charging habits, and adjusts to wear in the battery, brakes, tyres and suspension. Light personalisation should happen onboard, ideally without raw personal data leaving the car.
- The fleet learns collectively. One car’s rare encounter becomes training data that improves every car, and fleet-wide failure patterns are detected early, shifting quality management from reactive recalls to early intervention. Federated and privacy-preserving learning let models improve without centralising raw personal data, though this is technically harder and not yet standard practice.
- The ecosystem learns around the vehicle. Maps, traffic, charging availability, weather and infrastructure data flow in and out, allowing insurers, fleet operators, utilities and city systems to interact with the vehicle.
The resulting use cases include predictive maintenance, energy and battery intelligence, evolving driver assistance, conversational and agentic interfaces, and learning from near-misses as well as crashes. The car’s usefulness depends on how well it learns, how reliably it updates and how much the owner trusts it. Competitive advantage shifts to whoever best closes the loop: collecting the right data, learning from it responsibly and deploying improvements safely.
The winning question is no longer “What features should go into the car before launch?” It is “What learning loop, data foundation and governance model will allow every vehicle we sell to become safer, smarter and more valuable over time?”
AI, OTA updates and increasingly autonomous functions are opening new possibilities while also introducing new risks. How should automakers balance rapid innovation with functional safety, cybersecurity, data privacy and consumer trust as vehicles become increasingly software-driven?
Automakers should treat safety, cybersecurity, privacy and trust not as brakes on innovation but as design constraints that make rapid innovation scalable. The right balance is fast where risk is low, rigorous where risk is high: continuous delivery for customer-facing and non-safety features, and staged, evidence-based deployment for safety-relevant and autonomous functions.
Start with risk classification. Every AI feature, OTA release and autonomous function should be classified by safety criticality, data sensitivity, cyber exposure and reversibility. Threat analysis and risk assessment must happen before the architecture is frozen.
Use the right frameworks together.
- ISO 26262 governs functional safety against system faults.
- ISO 21448 (SOTIF) addresses hazards arising from performance limitations even when the system works as designed.
- ISO/PAS 8800 addresses the safety of AI in road vehicles.
- ISO/SAE 21434 and UN Regulation No. 155 set the cybersecurity engineering and management requirements.
- UN Regulation No. 156 and ISO 24089 govern software update management, including OTA.
Bound the AI. Higher-risk functions need rigorous validation, redundancy, degradation strategies and human handover. AI should automate what it can do reliably, but critical decisions need defined operating boundaries, monitoring and human oversight. AI-specific risks such as adversarial inputs, data poisoning, model drift and explainability gaps belong in the safety case. Treating them as purely an IT concern is a risk that organisations should actively guard against.
Make privacy and trust visible. Many consumers do not know what data their vehicle collects. Automakers should adopt a privacy-by-default posture: collect only the data needed for a defined purpose, and process sensitive data at the edge where possible. This is also what makes fleet learning socially acceptable. Trust is built by explaining data use, communicating updates honestly, naming capabilities accurately and demonstrating accountability after incidents.
India has established itself as a major global automotive engineering and technology hub. Where do you see the biggest opportunity for India to move beyond being an engineering and software services base and become a global leader in SDV and AI-defined vehicle innovation?
India’s biggest opportunity is to become a global originator of SDV and AI-defined vehicle platforms, not merely the world’s most efficient engineering delivery base. That means moving from building to someone else’s specification to owning the product, the platform and the intellectual property.
Why India. Three capabilities converge at a scale few markets can match:
- Engineering depth: powertrain, EV, battery, chassis, testing, homologation and manufacturing expertise.
- Digital talent density: embedded software, AI/ML, cloud, data, cybersecurity, simulation and product engineering.
- Demand-side diversity: dense cities, mixed traffic, two- and three-wheelers, price-sensitive customers, extreme climates and fragmented infrastructure. This is an ideal proving ground for robust, affordable, AI-enabled mobility.
Where to place the bets.
- A low-cost AI-defined platform for two- and three-wheelers and entry-level cars, with efficient on-device models and OTA capability at a fraction of premium-vehicle cost. The result is exportable to Southeast Asia, Africa, Latin America and the Middle East.
- Validation and scenario IP for complex traffic. Models validated in Indian conditions generalise well to other emerging markets, and owning the datasets, simulation environments and benchmarks is a strategic position.
- Indic multilingual in-car assistants that handle code-mixed speech, regional languages and local services such as UPI payments and charging.
- Products rather than services: simulation, MLOps, OTA management, cybersecurity monitoring and safety-assurance tooling that scale without linear headcount growth.
What must change.
- Move from billable hours to products, IP and outcome-based models.
- Build shared infrastructure: compute, open and privacy-safe datasets, simulation, and proving grounds equipped for ADAS testing.
- Clarify regulation: Bharat NCAP, ADAS and OTA pathways, and alignment with UN R155/R156 and global standards, so Indian-developed systems are exportable and trusted.
- Set practical data governance under the Digital Personal Data Protection Act, covering ownership, anonymisation and access for R&D.
- Back product startups through the scale-up gap with anchor customers and patient capital, not only seed funding.
Honest risks. The ecosystem is fragmented, with many capable players but few platforms of global scale. Chinese manufacturers are already targeting the same emerging markets with cost-optimised intelligent vehicles. And the profitable billable-hours model is hard to give up for riskier product investments.
Looking at the next five to ten years, what transformation do you believe will have the greatest impact on the automotive industry—and what should automotive leaders, tech companies and startups be doing today to prepare for that future
Over the next five to ten years, the greatest transformation will be the shift from vehicles as engineered hardware products to AI-enabled, software-defined mobility platforms operated as continuously evolving businesses. This changes not only what a car is, but how it is designed, manufactured, sold, serviced, insured, updated and monetised. The decisive question moves from “who builds the best car?” to “who owns the best combination of compute platform, data, models, validation capability and customer trust?”
Forecasts over this horizon are uncertain. Regulation, a high-profile safety incident, trade policy, chip supply or a breakthrough in model capability could shift timelines considerably. With that caveat, by the early 2030s it is reasonable to expect:
- Driver assistance as the baseline, with higher autonomy in bounded domains such as highways, logistics hubs, geofenced zones and commercial fleets.
- A small number of dominant compute and software platforms, from OEMs, big tech and Chinese stacks.
- Agentic in-car assistants as a standard expectation rather than a premium feature.
- Differentiation moving from hardware to behaviour, assistance quality and personalisation.
What automotive leaders should do now. Build one coherent SDV platform strategy (compute architecture, middleware, cloud connectivity, data foundation, OTA, cybersecurity and developer ecosystem) rather than funding disconnected digital initiatives. The organisation must be able to develop, validate, deploy, monitor and improve software after the vehicle is sold. That means product ownership, continuous integration, virtual validation, OTA governance, fleet telemetry, cybersecurity operations and post-launch analytics working as one system. Decide which layers you will own, and match development speed to the best new entrants.
What technology companies should do now. Earn a place in a safety-critical, long-lived ecosystem. Offer open, modular platforms that do not lock customers in, build compliance evidence into the product, invest in tooling for validation and simulation, and make AI efficient enough to run on low-cost automotive silicon.
What startups should do now. Focus on a specific bottleneck rather than a full stack, such as validation, battery analytics, driver monitoring or Indic voice interaction. Plan for long sales cycles, build compliance in from the start, own differentiated data or IP, and use regional advantages that global players serve poorly.
The industry’s next decade will be determined less by any single technology than by how well each player can build the capability to learn and change safely at scale. The companies most likely to thrive are those that prepare for adaptability rather than bet on one forecast, and that can respond quickly and credibly whichever way regulation, technology and customer behaviour turn out.
Connect with The Autonaut Media:
For editorial features, leadership interviews and media collaborations, reach us at pooja@theautonautmedia.com.
