A Chat With Miao Luo, Director Of Technology Strategy At Qt Group On The State Of OEM And Hardware Startups In The AI Age

As Original Equipment Manufacturers (OEMS) and hardware startups rush to slap AI onto their physical products out of fear of missing out, are they just building on a rotting software foundation? What does a ‘software-first mindset’ actually look like for a legacy hardware company trying to adopt AI?

Many OEMs and hardware startups are rushing to build advanced AI features on shaky foundations, but I wouldn’t say this rot is coming solely from the code. 

Particularly in the case of legacy OEMS, operations are typically stuck in siloes, with engineering, manufacturing, and supply chain teams all speaking different languages, making collaboration near impossible. And when organisations start trying to build AI into their physical products on top of these fragmented systems, instead of innovating, they inevitably end up increasing complexity, making the systems even more fragile. So chronic technical debt isn’t just an engineering problem; it’s a leadership failure. A byproduct of leadership treating software as simply an add-on to hardware, rather than the core value driver that it is, leaving teams spending most of their time maintaining the past instead of building the future. 

To adopt a truly ‘software-first mindset’, leaders need to reframe AI not as a way of keeping pace with trends, but as a byproduct of good software management. Because if you cannot manage the complexity of your existing codebase, you cannot safely manage or implement AI. 

When an AI hallucinates on a laptop, you get a bad spreadsheet; when an AI fails in a Software-Defined Vehicle or an industrial robot, people could get physically hurt. How do you bridge that trust gap, and what does the cybersecurity framework look like when AI intersects with heavy machinery?

The stakes are far higher with physical products like SDVs and industrial robots, so much so that we can’t rely on probabilities or wishful thinking. To bridge that trust gap fully, especially in such high-stakes environments, we need independent frameworks to monitor safety. 

In practice, it would work like this: an AI suggests an action, but it is verified by a safety-certified software component that must confirm that the output is within safe operational limits before any physical action is taken. Not only does this ensure physical safety, but it also creates an auditable trail for each decision, allowing human oversight into why the system is making the choices it does. 

On top of this, any AI that interacts with heavy machinery needs to be isolated from the safety-critical controls of the machine, so that in the event of any compromises, the underlying systems are protected. And cybersecurity measures like this need to extend beyond the production process throughout the full lifecycle, with automated quality testing to monitor and restrict AI capabilities in the event of any unsafe actions being flagged. 

As it stands today, companies are stuck in the hype cycle, wanting to use AI for everything because it is fast and easy to prototype. But fast is irrelevant if it isn’t safe. Trust isn’t built by making the smartest AI; it’s built by creating architectures so robust that even if the AI is wrong, the machine remains safe. 
 

The term ‘Software-Defined Vehicle’ (SDV) has become the gold standard for Western OEMs trying to pivot their strategy. Are we at a point where the industry is over-indexing on this label, or is it still a meaningful north star for traditional automakers?

Western OEMs are a little behind the times here, with many still benchmarking their strategies against that SDV label. In comparison, leading manufacturers in countries like China have already moved past the concept. The real leaders in this field don’t treat SDV as a marketing buzzword, but as a concept to be embedded across the manufacturing and product lifecycle.

That’s not because SDV is a bad north star – it’s that Western OEMs are held back internally by legacy siloes, fragmented ownership across both engineering and manufacturing, and slow decision-making. Holding the SDV label up as a goal won’t magically fix underlying structural issues. 

If they want to truly catch up with their Eastern counterparts, Western OEMs need to forget about the SDV label and turn their focus to redesigning their operating models. Software needs to be integrated into the manufacturing process from the very first product concept, rather than bolted on at the end. 

Historically, automotive innovation has been synonymous with hardware performance and engine capability. In a 2026 market where technology convergence is making many vehicles feel fundamentally similar under the hood, where does the real competitive battlefield lie for manufacturers?

In today’s market, most vehicles do arguably feel similar under the hood as hardware performance has converged. Instead, automotive manufacturers are turning to User Experience (UX) to differentiate themselves. Rather than competing for the best engine capability, manufacturers are building bespoke in-car differentiators that deliver more immediate differences for the users. While a driver might not notice the difference in engine capability right away, they will notice a change in the in-car experience. 

Imagine vehicles that feel fundamentally different depending on who is using them. Like simpler, more intuitive models for older drivers or highly configurable and advanced experiences for car enthusiasts. A tailored UX will make a vehicle stand out from a crowd of arguably similar hardware specs.