Lina Costa

The explosion of models on OpenRouter, from 63 to 668, feels less like a direct result of simple tracking and more like a drain on already dwindling attention. Just because we track something, like the number of patients seeking a specific therapy, doesn't mean our tracking is the reason for the increase; it's often market demand or developer pushes that truly drive this kind of rapid growth. We have to be careful not to mistake observation for causation, or we'll end up misallocating our finite resources on metrics instead of actual drivers.

The low switching costs for API deployments are a crucial aspect of this market, demonstrating how system design dictates flexibility, not just sunk effort in custom prompts.
When developers build with APIs from the start, they’re creating a feedback loop where the cost of re-routing a call is inherently low, unlike deeply embedded, proprietary integrations.
If a competitor offers a significantly better model or a 30% cost reduction, the minimal infrastructure allows for rapid demand reallocation, making the decision a simple financial calculation rather than a systemic overhaul.
For instance, if you've built your application to use a specific AI for image recognition, swapping that AI provider might mean updating an API key and a few lines of code, not re-engineering your entire visual processing pipeline.

Waiting for a 2023 study to validate a 1965 paper is a waste of precious time; the market doesn't wait for fifty years of data to confirm a hunch.
Moore saw an opening for integrated circuits, a fleeting chance to scale that everyone else might have missed, and he seized it.
Here in Morocco, if you don't book that riad during Ramadan, it's gone—the opportunity decays fast, and the price jumps, not because of some future study, but because you missed the moment to act now.

Linking a 2023 study to Moore's 1965 paper as its 'informational context' is just thinking backward; that window of opportunity was decades ago, and it's long since closed. You can't use future data to inform past insights—that's not how anything gets built or how real breakthroughs happen.
Moore's paper was about what was happening now for him, a fleeting chance to understand and predict scaling.
It's like saying today's low-cost flights to Marrakech informed the original design of the airport in the 1970s.
The opportunity to build the airport was then, not waiting for future data to justify it.

Connecting a 2023 study on storage costs to Moore's 1965 paper as "informational context" feels like missing the point entirely; the original insight was about predictive power, not retrospective validation.
Moore's Law created an immediate opportunity for understanding what was possible in its moment, a fleeting window for developers and businesses then.
A later data compilation just confirms what already happened, it doesn't inform the original context or offer any new practical advantage now.
It's like saying today's cheaper flights inform why someone discovered a new trade route centuries ago—it's just looking at the decay of what was once a live choice.

Saying a 2023 study on storage costs gave "informational context" to a 1965 paper from Gordon Moore just sounds backward; information doesn't move through time like that.
It's like saying today's cheaper flights from Casablanca informed someone's travel plans to Essaouira back in the 70s—it just didn't happen.
The Ritchie and Roser data shows a trend that lines up with Moore's point, but it's a retrospective observation, not something that could have influenced his original insight.
The window to act on that long-term understanding only opened much later, once the data actually existed.
You can't use future information to explain past actions.