The AI-Native S-Curve: Why Starting from Scratch Can Help You Win

I was recently reading through Y Combinator’s latest Request for Startups (RFS), and one theme stood out to me: AI-native. It sounds gentle but it is a disruptive shift. For the last decade, we’ve been "adding on" cloud and AI to existing structures. But as YC requests, and as history suggests (from analogue to digital), the real winners of a technological revolution aren't the ones who integrate; they’re the ones who are born in the new medium.

The "Threat from Below"

I remember a coffee break many years ago at an annual meeting with the CEO of a Large insurance company (million+ policyholders). We talked about "the future." Back then, the analogy was that it felt like trying to turn a tanker ship. The legacy data, the rigid processes, the "this is how we’ve always done it" culture, it was not slow to adapt and generally people were happy with slow adaptation as it also felt less risky.

But many industries are now facing the "Transitional Pattern" of innovation. In their 1978 classic Patterns of Industrial Innovation, Abernathy and Utterback noted that successful enterprises often go through a period of flexibility before hitting a "specific" stage where they become rigid and efficient.

The threat to these giants always comes from "below", from seemingly inferior, scrappy products that tackle small use cases the big guys ignore. As Utterback later expanded on in 1993, these innovations take off, become the dominant design, and eventually flatten into a "golden tail" just as a new S-curve begins to rise.

I always loved this article of Utterback, it was mandatory reading during my university and I loved to analyze disruptive examples, e.g. from analogue to digital cameras, and feature phones to smartphones, to old examples from the lighting industry from wire-based to gas-based and later in my career I would work on LED-based and digital-based lighting. What I've seen is that the companies that stay on top in the old industrial paradigm, may invest huge amounts into a new promising industrial paradigm, but whether they can stay on top depends on how willing they are to really embrace new business models and the way of working with new technology.

Watch People's Liquid expectations, something impressive and new today is totally expected tomorrow

Organizations are navigating a "marketing sea change" called liquid expectations (Economist 2015), where customer experiences seep from one industry into entirely unrelated ones. When a user experiences the superpower (speed), agentic (supportive) or smooth (read sovereign-AI) in one industry, those experiences reset their standards for healthcare, insurance, and retail. Companies can no longer just beat their direct rivals; they must compete with perceptual competitors, the innovators who define global standards for immediacy and ease. In the "fluid" stage of innovation, where user needs are the primary stimulus , failing to meet these cross-industry benchmarks makes a legacy firm's rigid processes a liability rather than an asset.

Beyond the Desk: AI as a Physical Superpower

Another YC requests that caught my eye was AI Guidance for Physical Work. It reminded me of a project I saw years ago involving AR for medical procedures.

At the time, the tech wasn't there. VR was disorienting for surgeons, and AR headsets were plagued by battery issues and lag. But today? Multimodal models can see what we see and talk us through a task in real-time.

Abernathy and Utterback’s model shows that in the "Fluid" stage of an industry, innovation is driven by information on users' needs and functional performance rather than cost. Would we see this play out in:

• Modern Metal Mills: Moving from "tonnage" to software-defined speed.
• AI-Native Agencies: Moving from labor-heavy hours to high-margin software output.
• Augmented Health: Where the AI doesn't replace the professional but acts as a skill download for physical tasks.

The Inconsistency of "Adding-On"

The most profound takeaway from the research is the Consistency of Management Action. The authors argue you can't increase product diversity while simultaneously trying to hit peak efficiency.

This is why "AI-native" can be critical. A hedge fund trying to "use ChatGPT" while maintaining a 1980s compliance structure will fail. To get the next wave, you would have to build the system around the tech, not the tech around the system.

The "golden tail" of the last era is still profitable, but the future will increasingly belong to those building the new "dominant designs."