Why scale creates new technological orders and what comes after AI models “History does not repeat itself, but it rhymes.” Making historical analogies in politics, the economy, or business is a tricky exercise. We look for them, we talk about them, we operate according to them, but they are more a property of our tendency towards reductionism than an inescapable truth. Here we will do the business-analogy game. I have one in mind that I find so practical, it might not even be an analogy, but a real coupling. And if this is the case, it gives us a useful lens for thinking about the future of AI. Technology’s Jumping Style I propose that technology leaps when a new order is created. A new order forms when a system organizes in a stable way that produces intrinsic value. One example from the natural world (there are many) is cells. When they organize with others to create a wing – it has formed a new order (organ) with emerging value properties. If this wing belongs to a bee and the bee forms a hive with its pals, this hive is a new order with its own value. Build a railway between two cities – easier to visit your parents. But if you build hundreds of thousands of miles all over a continent – space and time perception changes. Planning emerges. Build a telephone line between two cities – you can now call parents, not only visit. But if you build millions of miles between every house and every business all over the world, the concept of ‘turn-taking’ dissolves. Coordination emerges. Print millions of books, and you have put brains together across space and time. Knowledge persistence emerges. Large Investments Up The Scale The movie Contact has one of my favorite lines: “First rule in government spending: why build one when you can have two at twice the price.” So, what if I want more than two? How about four? A hundred? A cool million? Scaling has always been expensive and hard to justify before the emergent order appears. Mark Twain said, “A railroad is like a lie — you have to keep building to it to make it stand.” And the somewhat less famous Amory Lovins famously said, “We did the math, and it turned out that the industry had increased capacity … by 186,000 times,” in relation to the fiber network built in the late 1990s to 2000. True, these processes generated massive speculative losses and many failed companies. But we know where both infrastructures led eventually. The winners were the new order entities flowering from the rubble. Physical to Digital Order Building Remember books? One long-term consequence of knowledge persistence was the formalization of logic. It changed from being an Aristotelian reasoning ideal into an algebraic one. That, taken with mechanics and later chemistry, allowed the formation of transistors. Since the late 1950s, people have chosen to remember only the positive side of railroads, highways, electricity, telegraph, and telephone booms. A huge investment in building computing power generated a massive business impact. The ability to do more calculations, at an ever-growing rate, led to great prosperity. It progressed through the near monopoly of a few companies. Intel, Texas Instruments, and Samsung are investing heavily in R&D and building more chips – computing units – with exponentially growing abilities. These companies built fabs, and a little later, companies like TSMC and Samsung built “build for others” foundries. These “others” are companies such as NVIDIA, Apple, Google, and Amazon that specialize in designing the chips and accompanying software, not actually building them. foundries New Orders (Always?) Tramp the Old Two generations of order technology passed through this era. Chip builders – being oh-so-powerful – were huge companies amassing profits and monopoly status. Software companies, utilizing these chips (first order based on the computing skills of the chip), took over as leaders. Companies like Microsoft, Meta, and Google – building software, the first next order – passed the chip-makers in value. software The second order built over chips is knowledge design. Companies like OpenAI, Anthropic, Google, and xAI use hives of chips to create something new. What we call AI is, in practice, extremely powerful predictive systems. knowledge design Let’s recap what we established until now. New technology distribution is often measured in units. Books, km of rails, electricity lines, miles of fiber, number of transistors. And trillions of dollars. Scaling orders of magnitude is very expensive and very hard to justify before the next order appears. But luckily, we are human. Driven by greed, hope, and a must-go-up-with-the-assistance-of-bubbles financial system. What is the next technology order? What are the “units of AI” that need to be scaled? Which are the companies that will overtake the AI “block” providers? Introducing the Entity Let’s start with an assumption. The unit to be scaled already exists and has a name. And that is called an agent. The reason it makes sense is that an agent is a unit of knowledge. I don’t mean just in a subject matter. It can be knowledge in doing a task, in doing reasoning, even in “emulating” feelings (coming soon to the LLM near you. Do not mistake it for “having” feelings. We have put so much feeling into writing for thousands of years, it is surprisingly straightforward to approximate with current language models). An agent is a cell specialized in a task. It has the underlying DNA/LLM “knowledge” to be anything (and a rapidly improving “anything”), but it is “designed” to be really good in something specific. Complex AI applications today use dozens of agents in orchestration to achieve the users’ goals. Dozens of agents. Dozens What will happen if 100 agents work in orchestration? 1,000? A million? A billion? A new order will appear. For that order, there is no agreed-upon name, so I will suggest Entity. Entity Entity (proposed definition)
An entity is a coordinated system of specialized AI agents that collectively exhibits capabilities no single agent can sustain on its own.Unlike a single model or a linear workflow, an entity maintains parallel competencies, shared context and internal coordination mechanisms that allow it to operate continuously within a domain.
Entities are not defined by intelligence level, but by architecture: how agents are composed, how they communicate, how they persist and how they adapt over time.In this sense, an entity is to agents what an organism is to cells: not more intelligence per unit, but a higher order of organization. Entity (proposed definition) Entity (proposed definition) An entity is a coordinated system of specialized AI agents that collectively exhibits capabilities no single agent can sustain on its own.Unlike a single model or a linear workflow, an entity maintains parallel competencies, shared context and internal coordination mechanisms that allow it to operate continuously within a domain. An entity is a coordinated system of specialized AI agents that collectively exhibits capabilities no single agent can sustain on its own.Unlike a single model or a linear workflow, an entity maintains parallel competencies, shared context and internal coordination mechanisms that allow it to operate continuously within a domain. Entities are not defined by intelligence level, but by architecture: how agents are composed, how they communicate, how they persist and how they adapt over time.In this sense, an entity is to agents what an organism is to cells: not more intelligence per unit, but a higher order of organization. Entities are not defined by intelligence level, but by architecture: how agents are composed, how they communicate, how they persist and how they adapt over time.In this sense, an entity is to agents what an organism is to cells: not more intelligence per unit, but a higher order of organization. An entity can be knowledgeable across many tasks in parallel, which makes it effective in the world without requiring a single, monolithic intelligence. In practice, this can resemble general intelligence within a domain, without implying a universal or human-like mind. (I will not go into why not “one-agent-to-rule-them-all” and let you speculate about it alone, with your billions of specialized, non-stem cells). The Entity Design Market It is very early at this stage, but chips and LLMs have a lot in common. They are both general-purpose technologies and, as such, hard to measure and compare. Also, as such, they use “standard” benchmarks to compare in the market (“how long it takes to render a 4K movie” for CPUs and “grade in medical licensing exams” for LLMs). They are also optimized in ways that inflate benchmark performance. As entities grow in complexity, the bottleneck shifts from raw intelligence to coordination, reliability, and architectural discipline. Chips started with laying transistors on silicon, with expert chip designers taking over the hardware world. The next designers will plan the layout of agents and their relations to create entities. They will be intelligent designers. It went from Intel (chip builders) > to Nvidia (chip designers)  > to OpenAI (intelligence builders) > to the next generation companies (entity designers). Why are those entity designers needed? Why aren’t the “Intels” of the era enough to design “chips”, build a “computer”, and build useful “software?” Because of specialization. It is why Sharp used the transistors Texas Instruments built to make a calculator before TI did (though they thought about it…). It is why Microsoft is making operating systems and not Intel. And why Nvidia is better at designing chips than, well, Intel. There are outliers such as Google and Samsung that do what Intel’s Andy Grove pitched for – self-disruption before others do it for you. So, where do we go next? To companies building intelligence architectures so complex and so specific that the intelligence builders will not compete with them (in most categories). It doesn’t mean OpenAI will go away. TSMC and Samsung are super-strong. It also doesn’t mean the new designers will work in a silo. They will need tools. They will need services. Whole ecosystems will emerge. There will be software dedicated to designing, deploying, testing, and managing thousands or millions of agents within a single system. Similar to ASML in the chip-making industry. These entity designs will enable the next order down the line. What will that be? Multiple entities working in collaboration like a swarm of intelligent societies? Let’s leave that to sci-fi writers and focus on scaling from 10 agents to the first 1000.

The is an opinion piece based on the author’s POV and does not necessarily reflect the views of HackerNoon.

Amazon

Intel

Language is Civilization's Worst Inefficiency

C | B

Substack

Read My Stories

Founder

Co-Founder, Chief Architect

Too Long; Didn't Read

From Cells to the Future of AI and Everything in Between: Entity Formation

From Cells to the Future of AI and Everything in Between: Entity Formation

About Author

Comments

TOPICS

THIS ARTICLE WAS FEATURED IN

Related Stories

AI Knows Everything Except Why You’re Sad

I Jail-broke an AI

A Look at Use of API Gateways When Creating Complex Systems

Organizations as Complex Systems

When Machines Think: A Modern Candide on AI, AGI, and Human Meaning

Is it a Value Object or an Entity?

AI Knows Everything Except Why You’re Sad

I Jail-broke an AI

A Look at Use of API Gateways When Creating Complex Systems

Organizations as Complex Systems

When Machines Think: A Modern Candide on AI, AGI, and Human Meaning

Is it a Value Object or an Entity?

Light-Mode

Classic

Newspaper

Minty

Dark-Mode

Neon Noir

Minty

HN StartUps