Elias Stahl is the CEO & Co-Founder of HILOS, a software and production platform that reimagines how footwear is designed and made.
getty
In May 2023, I was about to close my company’s first meaningful priced round at a great price with amazing terms. Then the Memorial Day startup apocalypse came—a sharp shift in the venture funding environment—and cash was suddenly scarce. The founding partner at one of our investors was empathetic but blunt. The bar was being raised, and we had to sprint just to stay where we were.
In 2026, that’s how every startup feels, from the basement project to OpenAI and Anthropic. CEOs are hearing it from every side; from my CEO peer group to VC portfolio company summits, the conversations are about sprinting just to stand still, no longer to gain a competitive advantage but to just remain competitive.
Through that noise, there is a meaningful signal. Scott Penberthy, director of applied AI at Google, posted an analysis of 5,752 AI deployments across 2,385 public companies. It found that every layer of horizontal AI deployment, across infrastructure, operations and customer-facing services, had a negative ROI (ranging from -5% to -14%). Surprising? Not when horizontal AI is just the cost of competing.
This report also found that 73% of domain-specific AI deployments (vertical deployments) showed positive excess returns, as opposed to only 15% in horizontal deployments. Vertical deployments showed positive ROI where actual competitive advantage accrued, and across every use case, this was where constraints were specialized. Whether it's specific manufacturing lines, a surgery room, regulatory approval or drug development, highly specialized constraints allow companies to build their own data lakes, models and competitive advantage.
You’re not competing against another vendor or provider, but against what your customer can build themselves. In that challenge lies real meaning. How many businesses today are just wrappers over services they didn’t build, competing over UX, brand, relationships and integrations? When that is no longer competitive, we revert to the golden mean: deep expertise.
I see this firsthand with product design. While AI can generate beautiful 2D renders and even high-quality 3D meshes, it has never held a shoe. Perhaps the cost of sketching, concepting and 3D prototypes will all drop, but product design is about working within the constraints of material, manufacturing and usage. That’s where craft and expertise live. That’s where innovation is born.
The biggest obstacle organizations face during this shift is that horizontal AI is often easier to buy than vertical AI is to build. General-purpose tools can be deployed quickly, creating the appearance of progress. That is table stakes today. Lasting competitive advantage requires companies to invest in codifying institutional knowledge and creating feedback loops between experts and AI systems. That work is slower and often less visible, but it is also where the strongest returns emerge.
For executives evaluating their own AI strategies, the first step is to start asking where proprietary expertise already exists inside the business. The most valuable AI deployments are rarely the ones that touch every department equally. They are the ones that encode specialized knowledge, whether that's a design constraint, manufacturing process or testing expertise. AI becomes more defensible as it gets closer to the unique constraints that differentiate a company from its competitors.
AI is on course to change every industry. However, its greatest impact will come not from removing constraints, but from helping organizations understand and work within them more effectively. I've seen product designers become less dependent on manual trial-and-error and more reliant on living knowledge systems that capture generations of design decisions, testing outcomes and hard-earned expertise. The result is not less creativity, but faster learning and better judgment.
That pattern extends beyond design. In a study conducted by neuroscientist Vivienne Ming, the highest-performing results came from teams that actively combined human expertise with AI, outperforming both human-only groups and teams that simply relied on AI outputs. The advantage did not come from automation alone. It came from the productive friction between specialized knowledge, real-world constraints and AI-assisted exploration. As AI moves beyond generation and into real-world understanding, the conversation will shift from what these tools can produce to what they can help us innovate.
In future pieces, I’ll explore how constraint-aware systems are being built, what it takes to turn domain expertise into defensible AI infrastructure and how design, manufacturing and materials innovation are converging into a single, continuous workflow. The next wave of competitive advantage won’t come from using AI; it will come from embedding it where reality pushes back.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。