As I watched our Michelin IT & Digital teams experiment with spec-driven and AI-assisted development, I couldn't help but notice a striking and insightful parallel: it reminded me of the shift brought in physical production by additive manufacturing ("3D printing")—this moment when building things stopped being sequential and became intentional. What insights can we draw from this analogy?
As software engineers at Michelin, we're privileged to work in a company shaped by more than a century of manufacturing culture. While the parallels between a digital factory and a physical factory have their limits, they always offer an illuminating dialogue, helping each other anticipate the technological cycles that constantly reshape how we build, collaborate, and deliver.
When additive manufacturing hit the industrial world in the 1980s, it gradually rewrote the rules of how some complex things were designed and produced—incidentally, Michelin was among the pioneers through its joint venture AddUp, helping push metal additive manufacturing into real industrial territory.
It wasn't just a new tool—it was a new logic: intent over instruction, models over machining, automation over manual steps, and in some extent it redefined what it means to design, build, and scale.
To me, in many ways, AI-assisted software engineering seems to be the industry's "3D printing moment":
Intent-first replaces instruction-first. In 3D printing, you don't produce the part step by step—you define a high-resolution 3D model. In AI-driven engineering, you don't code line by line—you express a precise specification, and agents generate the system.
Delivery becomes fully automated. 3D printers turn a model into a finished object with no human intervention. Agentic pipelines aim at doing the same: specs, generation, review, tests, deployment.
Infinite variability becomes possible. Every printed part can be unique, without changing the machine. Every digital solution can be tailored on demand, at negligible marginal cost. Production-ready software becomes disposable; the cost of trying becomes lower than the cost of deciding—making product management a new possible bottleneck, as Andrew Ng recently pointed out.
Raw material is commoditized—the value shifts to the design. Filament or metal powder is cheap; design is what matters. Compute and tokens are commodities; the value lies in intention, orchestration, and architecture.
We may be entering a world where software becomes "additive": designed through intent, produced through automation, endlessly adaptable, and regenerable rather than maintained. This is a profound shift—and an exciting one.
