Here is my thesis: as AI tools reduce the effort required to produce software, the difficulty shifts from building to choosing the right problem and reaching a market. This is a reading drawn from my own practice, not a universal law of the industry.
I wrote this piece as part of my work at Ekioo. It is a personal reflection shaped by the products I build and the choices I face while building them.
The Commoditization of Software Infrastructure
In my work, I use AI to accelerate tasks such as authentication, payments, database migrations, API documentation, and CI/CD pipelines. The gain varies with the context, technical debt, and the level of oversight required.
That acceleration makes it possible to explore an idea faster. It guarantees neither product stability, security, nor adoption. Those qualities still require judgment, testing, and time.
My intuition is that this technical ease intensifies competition: when more people can test an idea, the mere existence of a feature becomes a less durable differentiator.
What Doesn't Commoditize
What seems to resist commoditization is a deep understanding of a specific problem, a trusted relationship with users, and the ability to make coherent product decisions. These are assets built over time.
I draw one working rule from this: distribution and trust deserve as much attention as technology. Copying a feature does not automatically reproduce an audience, a reputation, or a customer relationship.
The Token Vendor Observation
Another question guides my analysis: in an ecosystem built on usage-priced models, how much value goes to infrastructure providers, and how much remains with the products using them? The answer depends on costs, pricing, and each product's ability to create value of its own.
I therefore prefer to treat the "token vendor" idea as an economic hypothesis to test case by case, not as an automatic victory for providers.
The Practical Implications
Before building, I ask two questions: "Does anyone want to pay for this?" and "Why choose this product over another solution?"
In my practice, clarifying these answers early avoids confusing building speed with market validation. AI can accelerate a technical experiment; by itself, it does not provide evidence of demand.
My conclusion is deliberately cautious: technical speed remains useful, but it is not enough. I look for a more durable advantage in understanding the problem, building strong user relationships, and learning faster than a simple cycle of feature copying.
Top comments (0)