The original article is in Spanish, this translation to english is done by a locally hosted chatbot.
The digital economy is currently witnessing a stabilization of technical roles. While investment promoters rely on hyperbolic names for Machine Learning technology, calling it "Artificial Intelligence" or even "Autonomous Agents," especially those looking to profit from these pyramid schemes, there are others who are finding their rhythm not thinking in terms of intelligence or autonomy, but in uses that provide tailored solutions; this is a paradigm developed in the article ALLMs (Applied Large Language Models). In this framework, these technologies are not considered independent entities with agency, but treated as sophisticated text prediction engines that, alongside human intention, can produce highly useful results.
A key example of this technological settlement can be found in the industrialized market of intimacy. In her research Exposing OnlyFans: The Men Behind A $7.2 BILLION Lie, researcher Lo (Log Off With Lo) reveals that the $7.2 billion in revenue for the platform in 2025 is not driven by autonomous bots, but by teams of human "chatters" —the real hands behind the keyboard for 100% of high-earning creators.
High Precision Efficiency in the Sales Pipeline
In a previous article, I portrayed biological and mechanical efficiency as dependent on avoiding System Waste. Just as a cheetah only commits metabolic resources when the probability of catching prey is high, the OnlyFans Management (OFM) ecosystem focuses on extracting high-value income from whales (large clients: subscribers who spend between $45,000 and $60,000 in a single month).
The "product" here is not the content, but the feeling of being important to someone. It doesn't help them that users spend their time polishing and hammering prompts, feeling like they're arguing with an ACME robot.
To scale this, agencies have built a highly precise Sales Pipeline where each subscriber is treated as a Warm Lead (a customer close to closing the deal). Human labor is fundamentally more efficient for this task due to the extreme level of intimacy and continuity required to maintain the illusion.
Humans as Engines of Intention vs. Digital Automations
Digital automation is not conducive to intimacy-based decision-making. An autonomous digital agent carries a high risk of spitting out hallucinations or producing generic responses that break the parasocial illusion. In an industry like OnlyFans, such technical failures not only result in poor interaction but also risk closing the sales funnel for that prospect and their potential referrals, risking pushing the prospect permanently toward competitors.
Instead, humans are used more as Intent Engines. They provide the low-entropy, high-precision observation required to maintain the relationship. But how exactly is the flow here? The actual creator is limited to taking thousands of photos and videos in advance, often following a script from the agency, and stores them in a vault. The chatter (who could very well be a middle-aged person in Serbia working their shift) is the one who actually handles the chats and executes the sale.
As soon as the user interacts or signs up to receive information, the system triggers an automated alert marking them as a warm lead. The user receives a pre-recorded welcome video (which functions as a video sales letter) and the chatter launches a scripted opening line to establish rapport (emotional connection) and gauge the level of interest.
The human's intention comes into play to profile the victim. The chatter evaluates how quickly the user responds, whether they use emotional language, or ask personal questions.
Then the PPV Push (Pay-Per-View) occurs: once the prospect's profile is classified, the chatter starts a conversation by sending photos from the vault in sequence and at timed intervals designed to simulate that things are happening in real time in the way the user would like to see them.
They start with cheap content (around $15) to confirm that the user is willing to pay, and as the user engages, the chatter psychologically escalates the conversation, sexually provoking the user to raise the price of the locked messages up to $200 per photo or video. If the user doubts the identity, the chatter uses generative tools like ElevenLabs to create synthetic voice notes that break the objection.
This machinery is supported by three fundamental pillars:
Strategic Regulation: Humans act as the regulators of technology, ensuring that each token generated by the prediction engine (or cloned voice note) serves a specific strategic conversion goal.
Global Continuity: Unlike an individual creator, human teams work in 24-hour shifts to provide uninterrupted coverage in any time zone, ensuring that if a user writes at 2 a.m. feeling vulnerable, someone is always there to respond.
Architectural Memory: Agencies prioritize software like Infloww to track "emotional triggers" and personal details such as names, locations, and jobs. A human chatter uses this data in the CRM (customer relationship manager) to maintain a consistent narrative across different shifts, so that the user never notices they are talking to 5 or 10 different people.
An Ontological Frontier of this Technology
The Usage Horizon describes the final use of the final version of a technology that could be seen, witnessed, or detected from the current point of view. It is an ontological boundary.
Let's consider the usage horizon of a household hammer. Its final use is to drive in nails; its final version is an object that provides the perfect balance and torque to amplify human strength. Although we can calculate and design new materials to approach mathematical perfection, the standard hammer remains because it has already reached a point where it is "close enough" to that usage horizon. Since resources are limited, developing a tool beyond a certain state makes no economic sense. You could theoretically engineer a $10,000 hammer, but it won't be a thousand times better at driving nails than a $10 hammer. Even a $100 hammer relies more on brand recognition and marketing than on a 10x multiplier in real utility over the $10 hammer.
This limit is equally visible in more complex machinery, such as the automobile. Although development continues between everyday cars and Formula 1 vehicles, the central utility reached its horizon decades ago. This century's attempt to push cars beyond this horizon toward self-driving autonomous entities perfectly illustrates the asymptote: billions of dollars in capital have resulted in systems that still largely fail without human supervision and continue to pose legal, financial, health, and transportation risks when left alone.
Digitally Assisted Driving, used by the transportation industry for decades, has worked well enough for decades before tech bros forcibly inserted their own algorithms into the daily driving of the casual consumer.
The expenditure of resources required to achieve those last percentage points of true autonomy, which serve more for marketing and promises than for delivering results that change the daily driving scenario, aims to break the physical and economic constraints of the system.
The results are inferior to previous technologies, which are much easier to produce and handle, and the investors' urgency to create problems to keep selling is evident—not even solutions, but the promise of a solution.
Digital technologies are also subject to this limit. The Horizon of Use for chatbots is limited by the fact that supposed "Artificial Intelligence" or a "Singularity" is physically, materially, and chemically unattainable given current and foreseeable energy restrictions on Earth, upon which the hardware we build computation on operates.
Also considering the logistic reality of Landauer's Principle: the minimum energy required to erase a bit of information demands the generation of a minimum amount of heat, we can understand that current computational architecture is inefficient at handling this heat, and this supposed "artificial intelligence" is more of a residual byproduct of heat production.
As we try to scale toward "autonomous sentience," we crash head-on into the thermal and electrical ceilings of current silicon architecture (the materials and structures of current computing); achieving the "singularity" would require energy consumption and cooling infrastructure that exceed the sustainable capacity of any existing energy grid. It would be like wanting to create not only the technology for a miniature sun to burn—whose heating process also residually produces a machine for solving things that weren't problems and creating conveniences that are still unnecessary and produce nothing—but also having to develop the technology so that this miniature sun doesn't burn the entire planet.
Additionally, the material scarcity of high-purity neon for lithography and rare material elements like copper and electrical steel used for chip construction and high-density storage creates a physical flow that forces prioritizing efficiency over the infinite expansion that techno-oligarchs seek to impose by enslaving vulnerable countries and populations to mine and erode the planet's terrain.
This physical reality precedes the mass cancellations of data centers currently being reported, with nearly 50% of US projects planned for 2026 facing delays due to grid capacity failures and a shortage of custom-made power transformers that take 3 to 5 years to manufacture.
Just as a cargo ship cannot infinitely increase its speed due to the cubic relationship between speed and fuel consumption, all digital systems operate within a material asymptote where the next increment of "intelligence" costs more in environmental impact, market cannibalism, raw minerals, and gigawatts than it provides in marginal utility (and produces heat, I suppose).
That the final version of this technology is an autonomous agent is an impossibility; it is just another whip. The concept of the "singularity" is not a functional technical goal, it is a marketing mirage, a carrot dangled in front of investors and users.
We are left to bear the existence of these digital gods pushed forward with full force by shareholders trying to cram them everywhere, but physical constraints dictate that autonomous intelligence is not going to happen. That is why we suggest thinking about this in terms of ALLMs, instead of creating supposed intelligence.
And I call them digital gods because they also require us to pray to them; they require our use.
Co-Evolution as an Asymptote
I am not highlighting the OnlyFans management ecosystem as a cause of "AI" stagnation; rather, as an example of one of the consequences of it, of how a specific company exploits and cracks the whip on all fronts.
They use humans equipped with digital technologies because the technology reached its Horizon of Use a long time ago. The multi-billion dollar machinery simply operates at its best capacity, where text autocomplete optimally scales human intent.
And it is also something to observe in ALLMs; Large Language Models, even when applied, can be used in an exploitative ecosystem.
We are witnessing a stable system operating at the edge of this horizon:
The Operator: A Human Agent, providing empathy, strategic management, grounded intent, and final agency.
The ALLM: As a functional platform that enables and enhances the company's agency via the operator, for example, generating AI voice clones through tools like ElevenLabs to give a synthetic proof of authenticity, applying scripts and following pre-established user classification parameters, but still with a human operating all of it.
Operating on the Horizon
OnlyFans is a $7.2 billion testament to the fact that the most sophisticated use for ALLM technology is the enhancement of human-driven tasks.
Now, it is not necessarily the testimony of a company that uses it locally and modularly, conscious of its impact, but one that uses it on a large scale, indiscriminately, exploiting geopolitical frictions and differences in rights and legislation that protect the groups and possible positions of the participants.
Pushing technology beyond its practical ceiling burns finite minerals, electrical grids, and lands for crumbs of improvement that need marketing which is itself an achievement with more reach than the technology it promises.
The Horizon of Use marks the point where technological gains yield diminishing returns insufficient to justify the growing planetary cost; I am talking about trillions in rare earth extraction, grid overloads (US data centers delayed 50% by 2026), and gigawatt-scale energy consumption colliding with finite resources like neon and copper. For humans on a Planet with limited resources, chasing a digital slave with divine powers beyond this asymptote erodes ecosystems in exchange for marginal utility; it is like designing and producing $10,000 hammers that do not drive nails any better than a $10 one. It is unattainable because the thermodynamic, material, and economic costs make it irrational and unsustainable, long before we even start talking about the implications of what would happen if it were possible to do so. I wonder why anyone is still chasing that impossibility.
This supposed "Singularity" (that master state where machines will begin to think and create machines on their own, of their own design and with their own objectives) sells thanks to the Barnum Effect (in which vague and ambiguous propositions gain ground in individuals open, perhaps through manipulation, to them) and vague promises that inflate stock prices and scopes, in the same way that religious oaths achieve spiritual reaches without having in themselves proof, weight, or grace. It is a bubble built on potential efficiencies until the mathematics of resources kill returns, and as usual, the exploited workers pay the price.
Businesses run on short-term wins, not infinite growth. When energy and materials cap utility, companies that are clear about their product use hybrid, human-tool technologies; they do not go chasing mirages that break the ecosystems they depend on.
It is not just that we have failed to achieve "artificial intelligence" or the "singularity"; rather, we can simply recognize the Horizon of Use where autocomplete is truly planted, for what it is: a residual byproduct of heat generation. Large-scale waste.
By seeing these tools as co-evolutionary partners restricted by material reality instead of divine autonomous deities that for some reason are going to serve us as slaves, we leave behind the investors' marketing speculation, avoid falling into the trap of smoke-peddlers, and move toward a model of verified, high-utility execution.
I reiterate that the example is immense, but the proposal is small.
OnlyFans manages to exploit and squeeze its users and victims for billions, and it does so using chatbot algorithm technologies that autocomplete, generated as a byproduct of heat-generating machines, to automate tasks not as its primary logistic and action mechanism, nor its "purpose," but as a layer of functionality applied to human labor.
Fuentes
https://www.youtube.com/watch?v=G3VI_0XqwDY [Exposing OnlyFans: The Men Behind A $7.2 BILLION Lie, researcher Lo (Log Off With Lo) ]
https://en.wikipedia.org/wiki/Landauer%27s_principle [Landauer Principle] [5]
https://www.mdpi.com/1099-4300/26/5/423
https://quantumzeitgeist.com/landauer-limit-why-erasing-a-bit-generates-heat/
https://shalemag.com/why-ai-data-center-power-demand-is-reshaping-the-global-energy-grid/
https://www.utwente.nl/en/news/2019/1/44151/clever-timing-for-computers-producing-less-heat
https://en.wikipedia.org/wiki/Barnum_effect
