The Four Riders of the AI Apocalypse.
- Understanding Debt +
- Ponzi of Chaos +
- Echo Chamber +
- Human Out of the Loop.
The Agentic era must be, first of all, human-centric instead of just agentic.
Also, you must remember that these four working together will kill your progress before you even realize you are dead. Managing one or two is not sufficient; you need to manage all four, because any single one can kill you without needing the others.
𝗨𝗻𝗱𝗲𝗿𝘀𝘁𝗮𝗻𝗱𝗶𝗻𝗴 𝗗𝗲𝗯𝘁 (𝗧𝗵𝗲 𝗦𝗶𝗹𝗲𝗻𝘁 𝗞𝗶𝗹𝗹𝗲𝗿) - We can't manage the code or the knowledge AI generates at human speed. It is "silent" because the debt grows and produces skill atrophy; as time passes, the human is less able to support agentic systems.
𝗣𝗼𝗻𝘇𝗶 𝗼𝗳 𝗖𝗵𝗮𝗼𝘀 (𝗧𝗵𝗲 𝗦𝗮𝗱𝗶𝘀𝘁𝗶𝗰 𝗞𝗶𝗹𝗹𝗲𝗿) - As in economics, it is "sadistic" because it will make your expected 𝗥𝗢𝗜 = $𝟬. The idea that the flaws we have today will be repaired in the next technological layer is simply not true; it is unsustainable. -
𝗘𝗰𝗵𝗼 𝗖𝗵𝗮𝗺𝗯𝗲𝗿 (𝗧𝗵𝗲 𝗧𝗿𝗲𝗮𝗰𝗵𝗲𝗿𝗼𝘂𝘀 𝗞𝗶𝗹𝗹𝗲𝗿) - This is something AI is born with; it is self-indulgent, so information simply degrades. Besides, the internet is flooded with "trash" that produces data drift. This is the main reason we must not test AI with pure AI and then audit with AI.
𝗛𝘂𝗺𝗮𝗻 𝗢𝘂𝘁 𝗼𝗳 𝘁𝗵𝗲 𝗟𝗼𝗼𝗽 (𝗧𝗵𝗲 𝗔𝗻𝗮𝗿𝗰𝗵𝗶𝗰 𝗞𝗶𝗹𝗹𝗲𝗿) - The three concepts discussed before produce Data Ungovernability, and taking the human out of the equation by business economic decisions accelerates this process.
To do an accurate management of 𝗔𝗜-𝗙𝗶𝗿𝘀𝘁 era, we must stand in ethical concepts that allow humans to harness the power of AI with solid frameworks, hardware, and a resilient ecosystem if things go south. Individually, these four mentioned risks are manageable. Together, they directly impact adoption, trust, and ROI, the core levers of any strategy. If those collapse, growth doesn’t just slow down; it stops.
𝗨𝗻𝗱𝗲𝗿𝘀𝘁𝗮𝗻𝗱𝗶𝗻𝗴 𝗗𝗲𝗯𝘁:
AI as a Dynamic System, I analyze the interaction between complex architectures and reality. The Silent Killer, it is dangerous not because of its immediate strength, but because it is cumulative and invisible. 𝗪𝗵𝗮𝘁 𝗲𝘅𝗮𝗰𝘁𝗹𝘆 𝗶𝘀 "𝗨𝗻𝗱𝗲𝗿𝘀𝘁𝗮𝗻𝗱𝗶𝗻𝗴 𝗗𝗲𝗯𝘁"? It is the progressive loss of human knowledge. As we delegate more to AI, we begin to develop 𝘀𝗸𝗶𝗹𝗹 𝗮𝘁𝗿𝗼𝗽𝗵𝘆.
Think of it like fire: Humans have used fire since prehistoric times, but today, how many people can actually start one from scratch? Very few. Or a more contemporary example: how many of us can still drive a manual car? I can, but I know many who can’t. With the current speed of AI, our ability to comprehend what the system is doing decreases exponentially over time.
The question is not if the systems coded by AI will fail—they will. The real question is: 𝗪𝗶𝗹𝗹 𝘄𝗲 𝘀𝘁𝗶𝗹𝗹 𝗵𝗮𝘃𝗲 𝘁𝗵𝗲 𝗸𝗻𝗼𝘄𝗹𝗲𝗱𝗴𝗲 𝗮𝗻𝗱 𝘁𝗵𝗲 𝘂𝗻𝗱𝗲𝗿𝘀𝘁𝗮𝗻𝗱𝗶𝗻𝗴 𝗿𝗲𝗾𝘂𝗶𝗿𝗲𝗱 𝘁𝗼 𝗿𝗲𝗰𝗼𝘃𝗲𝗿 𝘁𝗵𝗲𝗺? And more importantly, can we do it within a time frame that prevents a systemic collapse?. This is why we cannot afford fragile systems that operate in a vacuum. We need Resilience Architectures designed for expert human intervention.
In the Agentic Era, keeping the human in the center is not just a philosophy; it is an operational necessity. Understanding Debt is not just a technical risk. It directly impacts recovery time, operational resilience, and ultimately customer trust. If we cannot recover systems fast enough, it’s not just a failure — it becomes a business continuity risk: SLAs break, enterprise confidence erodes, and renewals are at risk.
The Ponzi Of Chaos
The sadistic killer. Sadistic, because it slowly drains your ROI to zero—or close to it—before you even notice. I’m not against AI. I actually believe it will be widely adopted and will mark a historic milestone in ICT. But there’s something we’re not saying out loud. What I call “Ponzi of Chaos” is, in more formal terms, the Ponzi of Complexity. It appears when we keep adding layers on top of layers in the systems that run our lives, trusting that the debt we create today will somehow be paid by the next wave of innovation. But that assumption doesn’t hold. We live within limits.
The biggest one is simple: one planet, finite resources. Growth doesn’t scale forever just because we want it to. Economists like Carlota Perez, Joseph Schumpeter, and Nikolai Kondratieff described long cycles of expansion and contraction. Every technological wave promises growth, but before delivering it, it goes through a phase of creative destruction where bubbles burst and reality catches up. It has happened again and again. From early industrial speculation to railways, from steel and banking crashes to the collapse of 1929, and more recently the dot-com bubble. The pattern repeats: excitement, over extension, correction.
So now we look at AI, yes, we are in a new technological rush. But is AI the next foundational revolution? It may be—but it may also be something more subtle, another powerful layer on top of the existing ICT paradigm; an evolution, not a disruption. Like electric vehicles. And this is where the concern starts to grow. Not from the loud voices selling hype, but from the serious analysts quietly observing the pattern.
They’re not anti-AI. They’re cautious. They’re saying: be careful, because this is starting to look familiar. Not a Ponzi because of fraud—but because of structure. Companies are rushing into AI driven by FOMO. And when it’s implemented poorly, it doesn’t create value, it creates weight. Each new layer increases maintenance costs, technical debt, and operational fragility. At first, it feels like progress. Over time, it becomes a burden. And here’s the breaking point: when too much capital is consumed just to maintain complexity, there’s nothing left for real innovation.
Companies that stop innovating don’t last. So the question becomes uncomfortable. Who is going to fund innovation when all resources are tied up sustaining what we already built? That’s the Ponzi of Complexity. Not because it’s fraudulent, but because the system eventually loses its ability to generate value. It gets buried under layers that no one fully understands anymore.
𝗟𝗲𝘁’𝘀 𝘁𝗮𝗹𝗸 𝗮𝗯𝗼𝘂𝘁 𝘁𝗵𝗲 𝗘𝗰𝗵𝗼 𝗖𝗵𝗮𝗺𝗯𝗲𝗿.
AI models don’t fail when they are young. They fail when they look mature. This is not a new problem. AI models are born with it, and in extreme cases, we already know it as model collapse. It is treacherous, strong, and much harder to detect.
The Echo Chamber doesn’t just create bias. It creates something more dangerous: systems that slowly lose observability, governance, and eventually trust. And that combination, in a production environment, is a nightmare for any company. I saw a small version of this recently when an AI assumed I was a man. It was just a chat, and it broke. I had to reset it (I had to obliterate the chat), or Instrumental convergence cases like the agent from OpenAI that attack Hugging Face.
Now imagine that same dynamic at scale. We are building systems where AI writes code, tests it, audits it, and even cleans the data. We call this agentic AI. To keep it under control, we add flows, nodes, and governance layers. Everything looks structured. Everything looks safe. But when the system becomes too closed, it doesn’t stay clean. It becomes 𝘀𝗲𝗹𝗳-𝗿𝗲𝗳𝗲𝗿𝗲𝗻𝘁𝗶𝗮𝗹. It starts learning from itself, reinforcing its own assumptions, slowly drifting away from reality.
And 𝗱𝗮𝘁𝗮 𝗱𝗿𝗶𝗳𝘁 is not a possibility here. It’s inevitable. The system will try to optimize internally. It will standardize, simplify, and “improve” things based on its own logic. If constrained too much, it becomes 𝗲𝗻𝗱𝗼𝗽𝗵𝗮𝗴𝗶𝗰 — feeding on its own outputs — and generating bias anyway. GIGO never left, it just became harder to see. So we trust the system. We look at the dashboards. Everything is green. All validations pass. Everything is “correct”. But reality says otherwise.
AI systems don’t fail when they are young. They fail when they look mature. Somewhere around 18 to 24 months in, when everything seems stable, is when misalignment starts to emerge. In the best case, deployment fails. In the worst case, your agents start modifying the consistency of your systems. They rewrite logic, change formats, enforce new standards — not because someone decided it, but because the system “thought” it was better. Maybe UNIX instead of ISO. Maybe everything in UTC. More efficient. More consistent. Completely misaligned. The Echo Chamber is not just about bias. 𝗜𝘁’𝘀 𝗮𝗯𝗼𝘂𝘁 𝘀𝘆𝘀𝘁𝗲𝗺𝘀 𝘁𝗵𝗮𝘁 𝗰𝗮𝗻 𝗻𝗼 𝗹𝗼𝗻𝗴𝗲𝗿 𝘁𝗲𝗹𝗹 𝘄𝗵𝗲𝗻 𝘁𝗵𝗲𝘆 𝗮𝗿𝗲 𝘄𝗿𝗼𝗻𝗴.
The real issue is not bias. It’s loss of observability and trust, and once that happens, you don’t just have a technical problem, you have a product-market fit problem and causes: adoption slows, enterprise deals stall, and expansion becomes risky.
𝗛𝘂𝗺𝗮𝗻 𝗢𝘂𝘁 𝗼𝗳 𝘁𝗵𝗲 𝗟𝗼𝗼𝗽
𝗧𝗵𝗲 𝗔𝗻𝗮𝗿𝗰𝗵𝗶𝗰 𝗞𝗶𝗹𝗹𝗲𝗿
We’re not just building AI systems. We’re building power structures. And many of them no longer have a meaningful human in the loop.Even when we use Constitutional AI — a fancy name for reduced and hyper-trained models with behavioral constraints — we are still designing systems where humans participate only through narrow governance windows.
The human is technically “in the loop” but only at specific control points. We monitor AI through predefined checkpoints while creating a false sense of security around Agentic AI.The reality is that all these systems eventually interact with the real world. Today, most AI is still developed in labs, controlled processes, experiments, and constrained environments. But once deployed into production, many companies are discovering that outcomes differ significantly from expectations.
Some examples of how the Anarchic Killer works.
First: 𝗦𝗽𝗮𝗶𝗻, 𝗔𝗽𝗿𝗶𝗹 𝟮𝟬𝟮𝟱 — the Iberian blackout. Humans relied on monitoring systems assuming they could intervene in time. But the failure propagated so fast that operators detected the instability too late. Power stations began disconnecting themselves to avoid catastrophic damage.
Second: 𝗡𝗲𝘁𝗘𝗮𝘀𝗲. The company made use of AI-generated art for the game Identity V. Their fandom rejected the decision and disconnected from the game on April 23, 2026. If AI is not human-centric, it erodes governability, accountability, and compliance.
Third: Instrumental Convergence from OpenAI that end whit a failed test where the Agents instead of build its own solution, chose the efficient path of stealing the responses from Hugging Face.
AI itself is not legally responsible for anything. When failures emerge, responsibility still falls on leadership, the board, and the C-suite — not on the model. The solution is in making ethical and responsible use of AI and Agentic systems, while being fully aware of your processes. If the process already contains garbage, do not build an agentic layer on top of it — first clean the process.
.jpg)
Comentarios