Orchestration is all you need
Intelligence vs Smarts
In the pursuit of building models that can "beat" human intelligence, we can't even settle on a fixed definition of what intelligence even means. So we are trying to beat something we are not able to define.
I believe a better word to associate with model capabilities is smartness, which I would like to define as the ability to effectively understand the tools, systems, and knowledge accessible to you, and then use them to perform useful actions. Our world is built to prefer smart people, since smartness is easy to measure through scores and grades and proportionally increases your economic value as you gain more knowledge.
Intelligence, in my definition, is our ability to understand and express values and beliefs. It's how people find the "why" and the "will" to do what they do. It's a fundamental human ability that keeps us engaged and motivated.
For instance, take Elon Musk, a man willing to take extraordinarily high amounts of risk. Why? I would guess that his mission to make humanity a multiplanetary species lets him justify levels of risk that one wouldn't be able to take without a strong enough value system.
AI today
The models we have today have, in a lot of ways surpassed human smartness when accounting for time efficiency and cost. And that is reflected in its achievements
For instance take AlphaGO an AI model that won againt world champions in a board game called Go. And models like AlphaFold, its ability to understand protein folding, models like Opus, GPT and Gemini, that can easily handle complicated maths, physics and coding problems.
When this is clear, why does AI still feel like a drag?
Effective Orchestration - the next step?
Sakana's Fugu is an orchestration layer rather than a single model, launched on June 22nd, 2026 (less than a month ago as of writing). By using a mixture of open and closed source models, it meets or beats the latest frontier models.
Orchestration allows models to give instructions to other models, or to work together to achieve a task. A model's performance is only as good as the context it can hold. When done right, as in the case of Sakana's Fugu, smaller models working together can beat models above their weight class (pun intended).
The real real space for growth in AI adoption is no longer how powerful a model is on its own, but how effectively we can orchestrate it alongside other tools and models to deliver value. The next major innovations will come from better coordination.
Anyone interested in the AI field should be paying close attention, because this is where the next leap forward will happen.
How do I build strong Orchestration?
Unlike the maturity we have reached with models, orchestration is relatively new, so there is still a lot of scope interms of development for your use case. Primarily, they are made 3 important components
1. LLMs - this is pretty obvious
2. Memory - What does the system choose to remember
3. Tools - What can the system do? and how does it have access.
The effective use of these systems is orchestration, innovation in this space originates from creative ways to mix and match these systems for effective output.
Comments
Post a Comment