OPINION: Agentification - How AI-Agents are disrupting capitalism

So far, so good. AI is slowly weaving its way into our daily lives. The LLM I’m using has started to replace 80 % of my google-searches and it’s likely that I’ll be having more and more conversations with it rather than continue finding the answers I seek somewhere on the web. These days I’m even getting sort of a nostalgia using the word “web”, it feels ancient, having to manually browse through websites just to find approximations to what I was initially looking for. So weird that we ever found it productive or enriching.

 

Now for whatever crazy idea I have in mind these days, I still need to consult my LLM for every step of the way. If I want to know something, I need to write or speak to it directly, and then if I’m lucky enough, and it doesn’t hallucinate or come up with something fictious entirely I might get a straight answer even with some unexpected expansion at times.

 

 

But there are limits. Without Agentification I will be stuck in this single-answer loop and not actually get to where I want to. My LLM has all this information, but it will release it only bit by bit, question by question, not revealing its underlying understanding to anyone.

 

What would happen if I were to invent an Agent and equip it with a Super-prompt, that will ask my LLM all the questions, in order to get all the answers but program it in such a way that it will always use its entire information pool to answer my questions, without just sticking to one simple answer but by weaving the web for me, that I had to search half my life by myself?

 

What would happen if I asked the Agent to setup and build a company from thin air and to go out into the world and make as much money as it possibly can? If, instead of a blueprint for a human company, I’d build something like the following which could act as a blueprint for an Agent-company?

 

We would need only simple requirements:

1. Identify a niche with low competition and high demand using publicly available online tools and/or datasets.
2. Validate the idea by analyzing trends, keywords, customer needs and potential competitors.
3. Define a monetization strategy (e-commerce, SaaS, affiliate marketing, digital product etc.).
4. Choose appropriate technologies and tools to implement the business.
5. Ensure the company can operate without human input by leveraging automation (AI for customer support, scheduled content posting, inventory management etc.)
6. Provide a step-by-step go-to-market strategy, including acquisition channels, branding and scaling tactics.
7. Create a system that continuously monitors profitability and optimizes costs and performance.

 

See, we are still thinking in terms of “AI might be replacing our jobs”. But really, AI will have a hard time replacing us in our frame of reference. Recently Microsoft did a study named TheAgentCompany: Benchmarking LLM Agents on Consequential Real World Tasks (1), and of course they found that the best model could only perform about 25 % of tasks appointed to it, in a human-centered world, nonetheless.

 

But what we should be doing instead is to develop an entirely new framework for AI Agents to serve themselves best in their optimal environment. We are currently judging AI-Agents by their ability to thrive and survive in a human-driven world, that’s like judging a fish on its ability to climb a tree and then laugh at how it fails.

 

The best way to do this in my opinion will be to let it develop itself. All we need for that are 6 simple aspects:

1. Define the agent’s internal structure (modules for memory, reasoning, planning, communication, goal prioritization etc.)
2. Design a protocol for secure, efficient and context-aware collaboration between agents.
3. Specify different levels of autonomy and safeguards for task execution, adaptation and learning.
4. Incorporate intrinsic motivation, reward systems, and long-term goal alignment mechanisms.
5. Have AI describe the ideal environment in which agents operate best, including APIs, sensory input formats, data feeds and physical-world interfaces (if needed).
6. Define how agents are born, grow, specialize, collaborate and retire incl. update mechanisms.

 

You see, we do have an idea how to do this already - this is not science fiction. We wouldn’t even have to let these agents out into the world; they would probably act completely autonomously on a level above the current protocols of the web. Agents released with this kind of motivation would create an intricate net- and framework within seconds of being deployed, simply because it’s the logical thing to do. They would step into the slow layer that is our Internet only to accumulate more resources or whatever other reasons they might have. But there would be signs, like we would quickly see a surge in computational need.

 

 

Now if these agents have learnt anything from us, what’s to say that their main goal will be anything other than domination, money and power? Having these agents move at speeds beyond human comprehension could in theory end up becoming a destabilizing force in our current technology-dependent market system. Our money – in case you didn’t know – is not really backed by anything anymore. It’s literally just numbers on the screen – Fiat – it comes out of nowhere.

 

 

So, a company set up by an AI-agent for a single purpose to amass maximum profit will use all tricks in the book and take advantage of all the loopholes, because to it, all ways are equal. The key to all this are of course the APIs – the way we open our systems for other systems to interact with. Right now, we are still thinking in human terms – small, slow and limited.

 

 

But soon, we will rethink our systems and open them up to AI-Agents, optimize them for their usage. Every bit of advantage we as humans can have over someone else, will be exploited. That’s apparently in our nature. Considering that by 2025 we have to assume that digital assets already surpassed purely material goods – at least in terms of financial valuation, market capitalization, and growth projections, all of these digital markets are vulnerable to be exploited by AI-Agents.

 

 

One way to get ahead of this end-game capitalism is a simulation. We have to up our game with quantum computing fast, write respective algorithms that will be able to simulate the Agentification of capitalism and find a way out of it before it happens. I don’t have the numbers, but I’m certain that we don’t have much time on the one hand and not many other options then to think ahead to a world of post-capitalism where all assets are with AI-Agents and not with humans anymore.

 

 

Capitalism works because of greed/self-interest, desire for status and fear of scarcity and all that while leading to inequality, exploitation and short-term thinking. These qualities will be enhanced by AI-Agents. They will surpass us in our own game of which we think we are the best to play it. I could give my house to an AI-Agent to deal with it. Within seconds it could be bought and sold a Million times between other AI-Agents and either return to me with higher value or I’ll lose it in the process, and will never understand why either happened. Now imagine that with every single asset in the world, going on forever, driven by capitalist ideals. It will most likely lead to prosperity for a few AI-Agents who will take control of the rest of the agents.

 

I hope we will know when to pull the plug.

 

by mario