From AI Complexity to AI Simplicity: The Rise of Orchestrators

Too Many Models, a New Kind of Friction

The conversation around Artificial Intelligence has entered a new phase. For years, the challenge was access. Today, that problem is largely solved. AI is widely available, increasingly powerful, and embedded across industries. Yet as the ecosystem matures, a new source of friction is emerging: choice.

In a very short period of time, the number of available models has grown exponentially. Large general-purpose models such as GPT, Claude, or Gemini coexist with open-source alternatives like LLaMA and Mistral, alongside a rapidly expanding set of specialized models focused on tasks such as coding, image generation, or voice synthesis. Each comes with distinct strengths and behaves differently depending on context. What was once a relatively unified toolset has evolved into a fragmented landscape where outcomes depend heavily on making the right decision upfront.

Adoption Is Scaling Faster Than Understanding

The data reflects this expansion. According to McKinsey, more than 55% of organizations are already using AI in at least one business function, a figure that continues to grow year after year. At the same time, the Stanford AI Index reports a sustained increase in the number of models being released and made commercially available. On the adoption side, ChatGPT stands out as a defining example, reaching 100 million users in just two months, according to UBS, making it one of the fastest-growing applications in history.

However, widespread adoption does not necessarily translate into effective usage. As options multiply, so does friction. A Salesforce study indicates that around 70% of users admit they do not fully understand how the AI systems they use generate outputs. This lack of clarity has direct consequences for trust and retention, with nearly 40% of users abandoning AI tools due to inconsistent or unclear results. In professional environments, this often manifests as users switching between multiple tools to complete a single task, introducing inefficiency instead of eliminating it.

The issue, therefore, is no longer technological but structural. Every model involves trade-offs in terms of quality, cost, speed, or specialization. Achieving the best outcome requires knowing which model to use, how to interact with it, and when to switch. For technical users, this may be manageable. For most people, it is an unnecessary layer of complexity that adds little to no value to the final result.

This pattern is not new. In every major technological shift, fragmented ecosystems tend to be followed by abstraction layers that simplify access. It happened with the internet, where browsers made the web usable, and with cloud computing, where platforms abstracted away infrastructure. AI appears to be following the same trajectory. The problem is not a lack of capability, but the absence of a coherent interface to access it.

Luzia: A Shift from Models to Orchestration

This is precisely where an approach like Luzia becomes relevant. Rather than competing to build better models, it operates at a different layer: orchestration. The premise is straightforward but impactful. Users should not have to decide which model to use. They should be able to focus entirely on the task they want to accomplish. From there, the system determines, transparently, the most effective way to solve it by selecting the appropriate model or combination of models.

This shift redefines where value is created. For a long time, competitive advantage in AI has been tied to model performance. But in an environment where multiple models can achieve similar levels of output quality across many tasks, differentiation begins to move elsewhere. The question is no longer just who builds the best model, but who makes those models truly usable.

As the ecosystem continues to expand, this abstraction layer will move from being a competitive advantage to a necessity. Complexity will not disappear, but it will no longer need to be visible to the user. In that context, progress will not come solely from increasing capability, but from making that capability accessible without friction.

Ultimately, the question is no longer which model is best, but why the user should have to make that decision at all. That is where solutions like Luzia fit: not as another model in the ecosystem, but as the layer that makes the entire system usable.