Introduction
In 2023, an AI feature could be a product.
Upload an image and ask a question about it. Paste a document and receive a summary. Describe an idea and watch it turn into copy, code, or a plan. Each of these capabilities felt remarkable, and for a brief period, wrapping one of them in an interface was enough to launch a company. Access to intelligence was scarce, and scarcity created value.
In 2026, the situation has reversed. The same capabilities are built directly into the tools that millions of people already use every day. Image analysis, research, writing, coding, and reasoning are no longer rare abilities hidden behind specialized products. They are part of the default experience of using a modern AI assistant.
AI has not become less powerful. It has become more accessible. And when a technology becomes accessible to everyone, it stops being a product and starts being a commodity. The interesting question is what happens to value when that shift occurs.
Execution Is Becoming A Commodity
For most of modern history, intellectual work has been priced by time. Research took hours, so research was expensive. Writing documentation took hours, so documentation was expensive. Analysis, content, reports, prototypes: their value was tied, at least in part, to the effort required to produce them.
That link between time and execution is breaking.
Modern AI systems can complete a growing range of intellectual tasks in seconds rather than hours. They do not replace expertise, judgment, or accountability, but they dramatically reduce the cost of producing an acceptable result. A competent first draft, a working prototype, a reasonable analysis: these are no longer scarce.
This changes something fundamental about how businesses create value. When execution was expensive, simply being able to execute was a competitive advantage. Now that execution is cheap, it can no longer be the foundation of a business. Something else has to carry the weight.
What Looked Like A Product Is Becoming A Feature
Consider image analysis. In 2023, a tool that could look at a photo and describe it, extract text from it, or answer questions about it was a legitimate startup. Investors funded it. Customers paid for it. Today, that exact capability is native to every major AI assistant. What was once a product is now a checkbox.
Or consider the wave of applications generated with tools like Lovable. An individual can now describe an app in plain language and receive a working product in minutes: interface, database, authentication, deployment. The result is genuinely impressive. But it also means that the app itself, as an artifact, has lost most of its scarcity. If anyone can generate a functional application in an afternoon, owning a functional application is no longer a moat.
These capabilities still matter. They remain useful, and in many cases essential. But they are no longer rare, and rarity is what made them products in the first place. The question founders need to ask has changed. It is no longer "can AI do this?" The answer is almost always yes. The question is "why should this exist as a standalone product when the capability is available everywhere?"
The Risk Of Building Around The Model
Despite this shift, many AI products still follow the same recipe. Take a model. Add an interface. Add a few well-crafted prompts. Apply branding. Launch.
The problem is not that this approach fails immediately. It often works for a while. The problem is that it becomes increasingly difficult to defend. If users feel they are paying for ChatGPT outside of ChatGPT, the product is fragile by design.
Customers are not naive. They know the underlying models exist, they know those models improve every few months, and they know they can access them directly. Every model update narrows the gap between the platform and the product built on top of it. A business positioned in that gap is a business whose foundation shrinks with every release announcement.
Building around a capability that is rapidly becoming universal is not a strategy. It is a countdown.
Customers Buy Outcomes, Not Intelligence
Underneath all of this sits a simple economic point that is easy to forget in a market obsessed with capability: customers are not buying intelligence.
They rarely buy access to a model. They rarely buy an AI feature for its own sake. What they actually buy are outcomes. Faster decisions. Smoother experiences for their own customers. Time recovered. Confidence in the result.
The technology matters, but it is almost never the thing customers value most. It is the means, not the end. A restaurant does not pay for a language model; it pays for fewer missed reservations. A clinic does not pay for transcription; it pays for doctors who spend more time with patients and less time on paperwork.
As intelligence becomes abundant, the businesses that win will be the ones that are most precise about the outcomes they deliver, not the ones that showcase the most impressive capabilities.
When AI Stops Being The Product
This is where the thinking has to shift. If the model is no longer the product, what is?
The answer is the system built around it: what engineers increasingly call the harness. The harness is everything that turns raw intelligence into a reliable outcome. It is the integration with real data and real operations. It is the workflow that defines when the AI acts, what it is responsible for, and where a human stays accountable. It is the interface that makes the result usable, the guardrails that make it trustworthy, and the feedback loops that make it improve.
A restaurant booking platform does not create value because it contains a model. It creates value because it connects availability, reservations, confirmations, schedules, and operations into a single coherent experience. The model is one component inside that system, and often not the most differentiating one.
The harness is hard to copy precisely because it is not a capability. It is the accumulated understanding of a specific problem, a specific customer, and a specific context. Models are universal. Systems are particular. And particularity is where defensibility now lives.
Conclusion
The model is not the product. It never really was; it only looked that way while access was scarce.
Intelligence is becoming abundant, the way computing power and bandwidth became abundant before it. And like every technology that becomes infrastructure, it will stop being a competitive advantage and start being a baseline assumption.
That means value must move elsewhere: into the outcomes a business delivers, into the systems that make intelligence reliable, and into the judgment that decides where it belongs. AI features are not disappearing. They are becoming commodities. The businesses that thrive will not be the ones that build around the latest capability, but the ones that understand what remains valuable once everyone has it.
As a certain animated villain once put it: when everyone is super, no one will be. He was right about the economics. What he missed is that the power was never the point. The judgment behind it was. And that is exactly where value is moving now.
