Any project that integrates an LLM API inherently carries the risk of being absorbed by the LLM providers themselves. This applies whether you're building AI agents—whether coding agents, vertical-specialized agents, or general-purpose ones—or simply creating wrapper applications on top of the models. Ultimately, these projects can be swallowed by the very foundation model companies powering them. So far, we haven't seen a single LLM API-dependent project emerge as a durable, breakout star. Even Cursor has lost its lead in AI coding to Anthropic. The entire AI coding category is destined to be absorbed by the LLM giants. This dynamic also explains why we've yet to witness truly successful "AI applications." Last year's hyped experiments—apps that snap a photo to count calories or match you with a soulmate—were ultimately just clever wrappers around LLM APIs. They were flashes in the pan, nothing more. The projects that will thrive and evolve into massive, independent species within the AI ecosystem are those that both need LLMs and are needed by them—creating a relationship of mutual dependence, much like the one between LLMs and NVIDIA. One promising archetype for the next major independent species is a platform that doesn't rely on calling external LLMs at all, yet is indispensable to both LLMs and AI agents. Moltbook briefly captured attention as a potential example of this pattern. These transient but illuminating projects may represent the third paradigm in AI—following compute hardware and foundation models. In this model, the three paradigms remain distinct yet deeply interdependent. None is reduced to a one-way dependency. One-directional plays that simply consume LLM capabilities without offering something essential in return are inevitably commoditized or absorbed. Future third-paradigm AI species will share this same hallmark: they cannot be purely dependent. The LLM must need them just as much as they need the LLM.