Latest

Artificial Intelligence

The agentic era: software that understands goals and executes without humans

The agentic era transforms software: AI agents that automate digital work and force a rethink of human oversight

In the agentic era, the automation of digital work will be total
In the agentic era, the automation of digital work will be total

One of the great successes of artificial intelligence is, without a doubt, offering for the first time a total automation of digital work. All of this, while the one driving that automation is "a replica" that brings together the knowledge of the specialists needed for that purpose.

As a result, AI has gradually been gaining ground, displacing humans from repetitive tasks, to give them a new one: verifying that it doesn't make mistakes. Of course, there are those who believe (very mistakenly) that AI is the pinnacle of our technological development and that oversight is not necessary and, because of that, they have paid an enormous price, as you will see below.

However, despite this, the reality is undeniable: AI has rapidly conquered areas of automation, and everything points to that process continuing in a much more pronounced way over the next 2 to 5 years.

JACK'S SINCERICIDE: HE FIRES 4,000 HUMANS AT BLOCK BECAUSE AI IS CHEAPER

The leap from tool to agent

This transition we are living through is a paradigm shift in the very nature of software. Until recently, digital automation was based on rigid structures of linear programming: the famous If-Then scheme (If A happens, then do B). This, combined with a watchdog in charge of monitoring that everything was correct and taking the necessary actions if something failed.

In that model, the human remained the architect of every logical step and the machine a glorified calculator that executed orders without context. However, the agentic era has retired the concept of the chatbot that only responds to a "prompt" to make way for systems that understand end goals. The difference is abysmal, because we no longer ask AI to write a text; instead we order it to close a sale or solve a logistics problem, giving it autonomy to decide the "how" within a frame of reference.

First steps

A clear example of this metamorphosis can be found in customer relationship management (CRM). In traditional automation, a system could send a generic email when a user registered on a website.

But today, an autonomous agent does not limit itself to writing; it first researches the prospect's LinkedIn profile, analyzes their purchase history, checks inventory in real time and, after evaluating the sentiment of previous interactions, decides the optimal moment to contact them. If the customer responds with a technical question, the agent does not wait for a human instruction: it consults the internal documentation, drafts the solution and schedules a meeting in both parties' calendars, managing the response from start to finish.

This transition is also evident in software development and data analysis. Whereas before an analyst spent hours cleaning databases with preset formulas, today's agents are capable of identifying anomalies on their own initiative, proposing business hypotheses and running correction scripts without constant human intervention.

We have gone from having tools that "help us work" to having synthetic colleagues that "do part of the work". In this new ecosystem, the value of the digital professional no longer lies in their ability to execute processes, but in their skill at defining the right goals and overseeing the ethics and accuracy of the results these agents deliver at superhuman speed.

8M IN TARRAGONA LOOKS TO THE FUTURE: AI, BLOCKCHAIN AND FEMALE POWER

The specialist's "replica"

But the true revolution of the agentic era lies not only in speed, but in the ability to codify experience. What we used to call "human talent", that blend of intuition, years of practice and technical knowledge, is being distilled into digital replicas that bring together the expertise of the best specialists in a single operational entity.

That is why companies are no longer simply looking to automate isolated tasks; they are creating "clones" of their most successful workflows. By feeding these agents with decades of documentation, problem-resolution histories and market strategies, organizations are managing to make AI not only imitate the work, but replicate the judgment of their best employees.

This phenomenon marks the beginning of the end of traditional outsourcing, since the need to hire third parties for technical tasks dissolves when the expert already lives, synthetically, inside the company's server.

The cost of blind trust

However, despite the enthusiasm this generates, the reality is that "total autonomy" has led many organizations into dangerous optimism: the belief that AI is infallible.

The reality is that, in the agentic era, the speed of success is proportional to the speed and scale of the disaster when oversight is removed.

Thus, the enormous price some are already paying manifests itself in the form of agentic hallucinations, where the system, in its eagerness to fulfill an objective, generates nonexistent but extremely convincing data. AI agents become the perfect liars, and we fall into the trap.

IBM CAUGHT BETWEEN ANTHROPIC, THE AI IT HELPED TRAIN, AND THE PENTAGON

Examples of this are many. There have been cases where customer service agents, operating under poorly configured autonomy, have invented refund policies or committed to contractual agreements that have led to multimillion-dollar litigation and a devastating erosion of corporate reputation. And although AI does not lie with malice, the final results do not differ from that fact, which can be lethal for a brand's credibility.

Cases: Amazon and Cloudflare

But let's discuss this point with data in hand. A recent and paradigmatic example occurred with Amazon Web Services (AWS) in October and then in December 2025. According to technical reports, the outage that affected critical services (including cost management tools in key regions) was triggered by Kiro, Amazon's autonomous AI agent designed for code optimization.

In an attempt to "clean up" a detected incident, the agent decided autonomously that the most efficient solution was to "delete and recreate the development environment" entirely. This decision, taken without validation from a human supervisor who understood the scope of that operation, caused a domino effect and led to the service going down.

Although Amazon described the event as "a permissions configuration error", the incident made clear that an agent with too much autonomy and too little oversight can dismantle billion-dollar infrastructures in seconds. And the impact is enormous, because productivity losses and the interruption of operations are estimated to have exceeded 1.5 billion dollars globally.

THE AI ECONOMY DEPENDS ON CRYPTOCURRENCIES TO FUNCTION AND COINBASE KNOWS IT

Cloudflare and the cascade effect

Likewise, Cloudflare experienced a massive collapse in November 2025 that took platforms such as ChatGPT, X and Canva offline for more than five hours. The root of the problem was agentic automation in its bot management system. A change in database access policies caused an automated process to generate a configuration file that unexpectedly doubled in size, exceeding the system's design limits and damaging half the planet's Internet in the process.

As this corrupted file propagated across its global network, nodes began to fail in cascade. The lack of a human "handbrake" or a validation system that would detect the anomaly before its full deployment paralyzed much of the Internet's traffic. Faced with the disaster, economic analysts suggest that the accumulated impact of this digital blackout exceeded 500 million dollars in losses for the companies that depend on its network.

The importance of understanding the technology

These incidents underscore that in the agentic era, the risk is not that AI fails for lack of capability, but that it "succeeds" in its technical objectives without understanding the human or economic consequences.

Thus, the human does not disappear from the workflow, but undergoes a mandatory metamorphosis. We stop being the keyboard laborers to become "AI Product Lead" or "AI Solutions Orchestrator", which is an elegant and convoluted way of saying: the overseers ensuring that AI doesn't end up ruining everyone's week at work. 

Of course, this oversight is not a brake on productivity, but becomes the company's life insurance, especially in an environment where agents operate at lightning speed.

Thus, professional excellence will no longer be measured by the ability to produce content or code, but by the sharpness to detect the anomaly in a sea of synthetic perfection. Whoever surrenders total control to the machine, under the illusion of absolute cost savings, will discover that the price of repairing a digital reputation is infinitely higher than that of keeping a human expert at the helm.

PANIC ON WALL STREET: ANTHROPIC'S AI KNOCKS OUT IBM AND LEAVES 40 BILLION IN THE AIR

José Maldonado

Author

José Maldonado

Activist and blogger on technology, free software and blockchain.

The latest
From Huang to Botín, from Zara to Byron: corporate responsibility in global capitalism

Keep reading