The Smell of the Classroom
There is a specific smell in training rooms that never changes, whether the walls are adorned with 1997 calendars, heavy wooden furniture, or 2026 LED screens and plastic chairs. It is the smell of anticipation mixed with the fear of being left behind. For me, the classroom was not just a physical space with tiles and desks—it was the workshop where I learned the first lesson of the digital craft, attending various courses or presentations of new technologies. For nearly three decades, I have watched from the perspective of a student as entire generations of entrepreneurs crossed the threshold of that room, driven not by a passion for technological conquest, but by the primal instinct of economic survival. From the hum of the CRT monitors’ transformers to the sterile silence of data centres, the story is not about machines, but about the person trying to tame a new tool.
Looking back, I notice that every technological leap was met with the same deceptive promise of absolute simplicity. We were repeatedly told that the new interface would make everything easier, that the magic button would replace hours of apprenticeship, that the software would think for us. But the eye of the person who works with raw material—whether it is untreated leather or source code—knows that surface simplicity is often a trap. When you hide complexity behind a user-friendly interface, you do not eliminate the weight of the tool; you only move it from your hand into the shadow of the system, where you can no longer control it. True mastery does not lie in pressing a button, but in understanding what happens behind it.
In that room, people did not come to become engineers or programmers. They came with hands calloused from real businesses, minds burdened with profit and loss calculations, seeking a lever. Whether it was a floppy disk with data in FoxPro in 1997 or a subscription to an AI agent in 2026, their fundamental question remained merciless and pragmatic: How much does this tool cost? Who will operate it? And who will bear the burden when it fails? They understood, perhaps faster than the inventors, that technology is not a saviour, but a capricious employee, and the real cost is not the licence, but the dependency it creates.
Above and beyond business calculations, there is an unwritten law of history: any technological peak, no matter how civil it may seem at first, is inevitably absorbed by structures of power. From the servers that store our data to the colossal energy infrastructures that power algorithms, digital sovereignty has become the new battlefield. Those who believe they are adopting a simple productivity tool often discover, all too late, that they have become nodes in a global network of dependency. Legislation, data location, and human verification are not just bureaucracy; they are the last lines of defence for those who refuse to be mere passive consumers of their own economic destiny.
But before we talk about the cloud, sovereignty, and hallucinating algorithms, we must return to the origin of this relationship. We must remember the moment when the machine was not yet an abstract service, but a physical, heavy object that you could touch and had to learn step by step. The story I tell here is not a nostalgia for a bygone era, but a map for navigating the present. Because, although the screens have changed and commands have been replaced by prompts, the psychology of the person sitting in front of the unknown has remained exactly the same. Everything began with a silence, a cool room, and a cursor blinking on a green or absolutely black background.
Transparency Note: Article written with AI assistance (Qwen). Judgements, polemical positions, and perspectives belong entirely to the author.
First Encounter with the Green Screen: When You Know Nothing, and the Machine Knows It
There is a moment of absolute silence—a silence that is not a lack of noise, but a lack of direction—which marks the birth of any true relationship with an unknown tool. In 1997, in the computer lab, this silence had a physical weight. The air was heavy, smelling of heated plastic, burnt dust from the CRT monitors’ resistors, and barely suppressed human tension. The computer started with a deep hum, a transformer sound taking over the electrical load, and then, after a few seconds of waiting that felt like an eternity, it appeared. The green screen. Not a modern green of turned-off pixels, but a bright, active green, emitted by the cathode ray tube, over which hovered, solitary and relentless, a blinking white cursor.
That cursor was not an invitation; it was a challenge. It was an eye that watched you, waiting. There were no instructions on the screen, no colourful menus, no friendly hand to show you where to click. The machine knew you knew nothing, and this asymmetry of knowledge generated a primal fear, almost atavistic. The entire room kept their hands in their laps. Fingers were tense, withdrawn from the keyboard as if it were a scorching surface. The instructor’s floppy disk, a small black plastic box, square and fragile, was handled as if it were made of glass, with a reverence that betrayed the conviction that a single wrong move could destroy not just the data, but the very essence of the machine.
In those first hours, no one pressed a key without asking two, three times, looking at the instructor with wide eyes, seeking tacit approval before committing the act of typing. The fear of breaking something was infinitely stronger than the curiosity to discover. It was the fear of being the one who locked the system, the one who erased the future, the one who was not able to understand. But then came the first command. The instructor said: “Type DIR and press Enter.”
Fingers hit the mechanical keys with a dry, wooden-and-plastic clatter. Enter. And suddenly, the green screen was invaded by a cascade of white text. File names, extensions, dates, sizes. DIR filled the screen, revealing a hidden world, a structure, a hierarchy. It was the first revelation: the apparent chaos had an order. Then, the CLS command. A single Enter, and all that text disappeared instantly, leaving behind only the blinking cursor, untouched, in the top-left corner. That moment, that alternation between full and empty, between creation and destruction, generated the first real feeling of control. It was not an illusory control, but one based on direct causality: I had requested it, the machine had executed it.
Gradually, over those lab hours, a deep understanding began to coalesce in the students’ minds. The computer does not get angry. It does not get tired after a hundred wrong commands. It does not judge the intonation with which you speak to it or the fear in your fingers. It is an entity of absolute neutrality. It executes or reports an error. And even the error, that cryptic message like “Bad command or file name,” was not a scolding, but a simple statement of fact: what you requested does not exist or is not correctly formulated. This neutrality was, paradoxically, the greatest comfort. In a human world full of subjectivity, of nervous bosses and dissatisfied clients, the machine offered a playing field with clear, merciless, but fair rules. There, for the first time, a person was not judged for who they were, but only for the precision with which they formulated a request.
The Entrepreneurs in the Classroom — 1997: People Who Did Not Want to Learn MS-DOS, but to Understand What MS-DOS Could Change in Their Business
The classroom was not a temple of technology, but an intersection point of Romania’s transitional economic reality. In that room, in 1997, very different people gathered, united not by a passion for computing, but by an acute need for survival and adaptation. Among the students were small business owners, retailers, pension administrators, or production workshop managers. They did not come there to become computer operators. For many of them, this idea would have been almost offensive. They already had people who did the physical work; they came to see what it was all about, to understand the mechanism, to not be entirely dependent on an employee they could not control, and, most importantly, to not be deceived by them.
While the instructor explained the directory structure, these people did not write commands in their notebooks. Their notebooks were full of something else. They wrote business questions, translated into the language of the new reality: “How much does it cost to keep stock records on a computer?”, “Who can actually use it without breaking it?”, “What happens if the hard drive crashes—do I lose everything?” They looked at the screen with a different kind of attention, not technical, but economic. For them, every line of code was not an engineering achievement, but a problem to solve and a cost to evaluate. What problem does this solve, and what does it cost me if I do not have it?
They quickly understood, with the sharp intuition of people who had earned their living in the free market, that a computer with MS-DOS and a management program written in FoxPro could replace a folder with rods full of papers, two accounting ledgers, and, most importantly, half a typist’s workload. They were not particularly interested in the difference between DIR and CD or how to format a floppy disk. They were interested in whether there was someone who knew how to use it and whether that someone could be hired, trained, or, in the worst case, replaced if they became too arrogant. Their presence in the classroom was an early sign, an economic indicator of the fact that the computer was beginning to be perceived not as a state machine, a bureaucratic luxury, but as a business tool, a lever that could multiply human effort.
And yet, an interesting phenomenon occurred towards the end of the course. Many of these entrepreneurs left before it finished. They did not do so out of lack of interest, boredom, or the belief that “it’s not for me.” They left because they had obtained exactly what they had come to find out. The answer to their fundamental question had been found: Yes, the computer can do this; yes, there are solutions; yes, it costs X lei per month with an operator. Once the economic equation was clear, their physical presence in the course became an opportunity cost too great. Their time was too valuable to spend learning command syntax. They delegated. This early departure was not a failure of the course, but a validation of their pragmatism. They left the room on the shoulders of the others, those who stayed—the ones for whom the computer was not just a tool to delegate, but an extension of their own mind.
The One Who Stays After Class: The Difference Between Two Weeks and Six Months
The course ended. The classroom doors closed. The noise of chairs being pulled out and voices commenting on the latest news faded. People left with their certificates in their pockets, satisfied that they had ticked off a new skill on their list. But I stayed. Not because I was forced to, but because something in the emptiness of that silent room called to me. The teacher, a man tired of so much repetition, said nothing. He saw that I did not break anything, that I was there with a purpose, that I touched the keyboard not with fear, but with deliberate curiosity. The agreement was tacit. He allowed me to stay after hours, as long as I respected the working schedule limits. This was my first real lesson in autonomy: no one was guiding me anymore.
The days repeated themselves, turning into a ritual. In the morning, I turned on the computer. I pressed the big button, heard the hum, waited for the boot sequence, saw the memory test. In the evening, I turned it off. This repetition was not monotonous; it was the fundamental construction of familiarity. The machine became as familiar as my own desk, my own workspace. I learned to navigate through directories without thinking about the intermediate steps. I discovered, through trial and error, that the same thing could be done in multiple ways: you could copy a file with COPY, move it with MOVE, rename it with REN. This discovery of redundancy was a moment of liberation: I understood that I was not a slave to a single path, but an explorer who could choose the route.
I made mistakes I could not have made in the two weeks of supervised training. I deleted the wrong directory. I formatted the floppy disk with the data I was supposed to keep. I modified the AUTOEXEC.BAT file, and the computer no longer started correctly. And that was precisely why I learned more. In the controlled environment of the course, mistakes were corrected before they had consequences. In the solitude of the classroom, the mistake was mine, the consequences were mine, and the solution had to come from me as well. Those moments of panic, followed by the feverish search for a solution, the repair of the system, were the ones that forged true competence.
The difference between a trainee and a real operator is not measured in the number of commands memorised, but in the hours no one has scheduled, in the nights spent trying to understand why a certain program would not load, in the quiet satisfaction of making a system work again. Those unscheduled hours were the craftsman’s workshop. There, far from the teacher’s eyes and the pressure of colleagues, intuition was built. I learned to listen to the sound of the hard drive and know whether it was reading or struggling. I learned to recognise error messages not as sentences, but as clues. Staying after class was the act of transitioning from a consumer of information to a master of the tool.
When the Computer Becomes an Instrument: Conclusion Without Rhetoric
Time passed, and the relationship with the machine transformed. I no longer read the commands before typing them; I knew them. My fingers typed them on their own, like a pianist who no longer looks at the keys. I no longer panicked when something did not work. Instead of freezing, I read the error and thought. “Access denied” meant I did not have permission or the file was read-only. “File not found” meant I was in the wrong directory or had mistyped a character. Panic was replaced by a cold, clinical analysis. The computer was not a capricious sorcerer, but a logical mechanism that I had learned to decode.
I discovered the files that controlled the computer before I even touched anything. CONFIG.SYS and AUTOEXEC.BAT became my reference books. The first time I modified them, adding a line to load a mouse driver or free up conventional memory, was a new, almost sacred feeling. I understood that I had access to the soul of the machine, to the way it prepared itself to serve me. I was no longer just a user running applications; I was a configurator of the environment in which those applications lived.
During this period, viruses were not an abstract metaphor, a concept from Hollywood movies. They were a physical, tangible threat. They came on floppy disks borrowed from a friend, from a magazine with shareware software. I learned to be careful, to check every floppy disk with an antivirus that scanned slowly, line by line. I had worked six months to understand the system, I had important files, configurations I had made myself. I had something to lose. This vulnerability made me more responsible, more attentive to detail, more respectful of data integrity.
And then came the moment when I received the certificate. “IBM PC Operator”. A piece of paper with a stamp, signed by the centre’s director. Those who had left after two weeks had it too—those who barely knew how to turn on the monitor. Those who had stayed until the end, but only to tick off their attendance, had it too. I knew what it said—a formal title, a bureaucratic recognition. And they did not. They thought the paper made them operators. I knew that only the hours spent in front of the green screen, the repaired mistakes, the understood configuration files, made someone an operator. The paper was just proof for others; the knowledge was proof for myself. This is the difference between a tool and a tool left in a corner: the tool is the extension of the hand that has honed it.
AI Presentation in 2026: Same Cinema, Different Film, Same Room
The year is 2026. The conference room has been renovated. The tiles have been replaced, the walls painted in neutral colours, and in place of the old CRT monitors, there is now a 4K projector and a high-speed Wi-Fi network. But the basic structure of the room, the geometry of the space, is the same. And, surprisingly, the human dynamics are also the same. The room is full of entrepreneurs. No one comes to become an “AI operator”. Everyone comes to see what this AI thing is all about, to not get left behind, to not be dependent on a technology they do not understand.
The hype has done its job before the presentation even begins. “Artificial intelligence” sounds in their minds like a magic button, like an employee who never sleeps, never asks for leave, never goes on strike—a solution that manages everything, from accounting to marketing. Expectations are set at the level of a miracle. The presentation begins. And the first feeling that settles in the room is one of deep familiarity, a déjà vu that spans three decades: the screen does not show what you expected.
It is not a magic button. It is not simple. It is an interface, a text box that asks you to know what to ask, how to ask, and how much to ask. The AI of 2026 is the MS-DOS of 1997: powerful, real, useful—but mute in the face of those who do not know how to speak to it. Features multiply on the screen: text generation, image creation, complex data analysis, automation, autonomous agents that execute tasks. Each seems simple on the surface, demonstrated by an enthusiastic presenter who achieves spectacular results in a few seconds. But each hides a depth that is not revealed in the demo.
The price seems small at first glance. “Just a few dollars a month for a basic subscription,” says the presenter. A reasonable cost, accessible to any small business. And that is precisely the trap. An agent left to run unsupervised, a poorly configured automation that enters an infinite loop, a query that exponentially multiplies towards an external database—and the budget for a single day exceeds the budget allocated for an entire month. The complexity is hidden beneath the simplicity of the interface.
Then comes the coffee break. And this is where the real conversation begins. Not during the presentation, but at the buffet, with a cup of coffee in hand, where the masks of politeness fall. “Well, and who operates it? Because I don’t have time,” says a logistics company owner. “I have employees; I don’t have time to learn this new DOS,” adds another. “If I need someone to operate it properly, how much does a person with a project manager role cost, someone who understands both the business and the AI?”
Someone takes out their phone, quickly searches a job site. A brief silence falls over the room as the result is read aloud: “€2,500 to €4,000 per month for a project manager with proven experience in AI implementation.”
The silence grows heavy. “So what have we solved?” someone asks, with a note of disappointment and, at the same time, sudden clarity. “We took out a €20 subscription and need a €3,000 person to use it?”
This is the right question. And no one in the official presentation had anticipated it, because presentations are made to sell the dream, not to expose the real cost of implementation. The real cost of AI is not the monthly subscription to the software platform. The real cost is the person who knows how to drive it, verify its results, supervise it, and, most importantly, stop it when it goes off the rails and starts generating hallucinations or spending money without sense.
The entrepreneurs of 1997 were doing exactly the same calculation, in the same room. The computer cost 400 German marks—a considerable sum at the time. But someone had to operate it. And that someone, the qualified operator, cost more per month in time, salary, and dependency than the machine itself. Some left halfway through the 2026 presentation. They were not disappointed by the technology, but correctly, pragmatically concerned. They had realised that they had not found a replacement for thinking work, but a tool that demanded even sharper thinking.
Those who stay in the room until the end understand a fundamental truth: here too, as in 1997, there is a distance between a simple tool and a precision instrument. That distance is not covered by the software manufacturer, not filled by the paid subscription. That distance is built by you, through hours of study, through mistakes, through a deep understanding of the system’s limits and possibilities.
The certificate has a different name now. It no longer says “IBM PC Operator”. Now it says “Prompt Engineer,”“AI Consultant,” or “Digital Transformation Specialist.” It sounds more modern, more expensive, more impressive. But the fundamental question remains exactly the same as it was in 1997, when you stared at the cursor blinking on the green screen: Do you know what it says? Or is it just another piece of paper hiding, under a pompous title, the same lack of understanding of the craft?
Technology changes, interfaces become friendlier, algorithms become more complex. But the fundamental law of the tool remains unchanged: it does not replace the craftsman—it demands a better craftsman. And those who understand this, those who stay after the room empties, are the ones who, in the end, will transform not just their business, but the way the world works. The rest will be left with the €20 subscription and the illusion that they have bought the future, while the future waits, patiently, for someone who knows how to speak to it.
Windows 3.11 — The Graphical Revolution and the Price of Dependency
The installation of Windows 3.11 came as a liberation. After months spent in front of the green screen, typing commands that had to be memorised exactly, the appearance of the graphical interface was met with almost childlike enthusiasm. Everyone said it was good, that it was much easier, that you no longer had to memorise cryptic syntax. The mouse had become an extension of the hand, windows opened with a click, and the buttons seemed to invite you to press them. It was a democracy of access: anyone could now use the computer without being initiated into the secrets of the command line.
But this joy was not unqualified. The graphical interface hid the complexity, but did not eliminate it. And worse, with the ease of access, viruses appeared. It was no longer enough to know how to navigate through directories; now you had to be careful with every floppy disk, every downloaded file, every program run. The virus was not just a system error; it was an intruder, a parasite that hid behind friendly icons. Windows 3.11 gave you the illusion of total control, but in reality exposed you to risks you did not fully understand.
This 1997 experience is now waiting, projected onto the Artificial Intelligence of 2026. Just as Windows 3.11 seemed to simplify everything, hiding the complexity of the operating system beneath a user-friendly interface, AI also promises a deceptive simplicity. But just like back then, this simplicity comes with a hidden price: dependency on an infrastructure you do not control, on algorithms you do not understand, and on risks you cannot anticipate.
What we did not understand then, but is clear now, is that global shifts in the ownership of authority depend directly on digital infrastructure and the energy resources required to maintain it. Windows 3.11 was not just software; it was a step towards the centralisation of power in the hands of those who controlled the standards, licences, and updates. Similarly, the AI of 2026 is not just a tool; it is a system that consumes colossal energy resources, requires huge data centres, and, ultimately, consolidates the power of those who own this infrastructure.
Any technological peak brings with it temporary supremacy. Whoever masters the cutting-edge technology controls the flows of information, economies, and, ultimately, political decisions. Windows 3.11 was such a peak in the 1990s, and Microsoft became a giant not just because it made a good product, but because it understood that technology is an instrument of power.
But naturally, in history, this technological peak was absorbed by the state with a role of military supremacy. What began as a civilian innovation, as a tool for productivity and communication, quickly became a strategic asset. Governments understood that whoever controls the digital infrastructure controls the future. Wars are no longer fought only on battlefields, but also in cyberspace, and AI is the new battlefield.
In 1997, when you installed Windows 3.11, you did not think you were participating, even at a microscopic level, in a global reconfiguration of power. You thought you were making your work easier, becoming more efficient. But each installation, each licence, each update consolidated a system that, in the end, served the interests of the most powerful.
And now, in 2026, when the entrepreneurs in the training room look hopefully towards AI, the same pattern is repeating. The enthusiasm for the simple interface, for the promise of automation, hides the harsh reality of dependency on infrastructures controlled by tech giants and the states that support them. Just as in 1997, those who adopt technology without understanding the strategic implications involuntarily become pieces in a power game they do not control.
The viruses in Windows 3.11 were an early warning: ease of access comes with vulnerabilities. Today, the vulnerabilities are much deeper. It is no longer just about corrupted files, but about existential dependency on systems you cannot operate alone, on algorithms that make decisions on your behalf, and on infrastructures that can be shut down at any time by those who own them.
The Windows 3.11 experience teaches us that every technological leap is, at the same time, a liberation and a captivity. It frees you from technical complexity, but captivates you in a system of dependency. And, historically, this system is inevitably absorbed by existing power structures, which transform it from a tool of the individual into an instrument of state and military supremacy.
Those who stay after class, those who understand not just how to use the tool, but who controls it and for what purpose, are the only ones who can navigate this reality without becoming mere subjects of it. The rest will continue to install updates, pay subscriptions, and believe they are free, while their authority gradually dilutes into digital infrastructures they do not own and do not understand.
Legislation as the New Command Line: Data, Sovereignty, and Human Verification
If in 1997 the only “law” governing the computer was the binary logic of the processor, in 2026, over the software layer and the technological hype, a dense, invisible but extremely rigid mesh has settled: legislation. The state, which historically absorbs technological peaks to consolidate its supremacy, is no longer content to own the infrastructure; it now dictates the rules of engagement with it. For the entrepreneur in the training room, the question is no longer just “What can this tool do?”, but “What does the law allow me to do with it?”
The First Barrier: Data Types
In 1997, data was tangible. You held it physically in your hand in the form of a 3.5-inch floppy disk. You knew exactly where it was, who touched it, and who could copy it. Today, data is fluid, but legislation (from GDPR to European AI Acts) classifies it with surgical severity. Not all data is equal. Customer contact details, the company’s financial history, production recipes, or pricing strategies carry immense legal weight. Uploading this data to a public AI model in the cloud is the digital equivalent of leaving the cash register and the company stamp on a park bench. Legislation draws clear boundaries: there is data that can be processed externally and data that must never leave the company’s security perimeter. Ignorance in this area is no longer an excuse, but a guaranteed fine.
The Second Obligation: Server Location
In the MS-DOS era, the server was the hard drive under the desk. You could unplug it, pick it up, and leave with it. Control was absolute. In 2026, when you press “Enter” in an AI interface, your request travels through fibre-optic cables to a data centre that could be in Frankfurt, Dublin, or California. European legislation attempts to impose the concept of digital sovereignty, requiring providers to declare where information physically resides. The pragmatic entrepreneur of 2026 is forced to ask the same question they asked in 1997 about stock: “Where is my merchandise?” If your company’s data is processed on a server outside the legal jurisdiction that protects you, you have lost control. You have not just outsourced a task, but a part of your business’s sovereignty.
This awareness has generated a natural reaction, a direct echo of the craftsman’s mentality who wants their tools in their own workshop: local AI processing. Just as some entrepreneurs in the ’90s refused to outsource their accounting, in 2026 the movement towards open-source AI models running on their own infrastructure is emerging. A powerful server, placed in the room next to the office, hosting a language model capable of running offline. It is a solution that requires serious investment in hardware and technical maintenance skills, just like in the days of configuring AUTOEXEC.BAT files. But it offers something no cloud subscription can: absolute privacy and total control. The data does not go anywhere. Processing happens locally. It is the choice of those who understand that true technological independence has a cost, but that the lack of it has a much higher price.
The Third Imperative: Human Verification of Outputs
Regardless of whether AI runs in the cloud or locally, legislation and common sense impose one final, absolute condition: human verification of outputs. The concept of “human in the loop” has moved from engineering jargon to a legal obligation and a survival necessity. Artificial Intelligence, no matter how sophisticated it may seem, is a probabilistic engine, not a truth engine. It can generate a contract that is impeccable in form but contains a clause that nullifies your guarantee. It can create a convincing financial analysis based on a numerical hallucination.
The law states clearly: the final responsibility for a decision made based on AI-generated output rests with the legal or natural person using it. You cannot sue an algorithm. You cannot claim damages from a neural network. In 1997, if you made a mistake in a delete command, you lost a file. In 2026, if you blindly accept AI output without verification, you can lose a contract, violate competition law, or destroy a reputation built over decades.
Human verification is not just a tick in a compliance process. It is the supreme act of the digital craftsman. It is the moment when the experienced eye, the one that learned to read errors on the green screen, looks over the machine’s result and says: “This is not right.” The “AI Operator” or “Digital Transformation Specialist” certificate does not absolve you of this responsibility; on the contrary, it amplifies it. Those who left the 1997 classroom thinking the paper excused them from the work were just as mistaken as those who leave the 2026 presentations believing AI is an autonomous employee.
In the end, legislation is not an enemy of innovation, but the guardian that forces you not to forget who is actually in charge. The machine executes, but the human is responsible. And in the digital workshop of the 21st century, just as in the tanner’s workshop of old, the quality and safety of the final product ultimately depend on the hand and judgement of the one who wields the tool.
Selective Online Bibliography
- Doctorow, Cory — “Enshittification: How Platforms Decay” (2023), published on the personal blog Pluralistic and developed in the book The Shakedown (2024). Fundamental source for understanding the degradation of technology platforms and the pitfalls of cheap subscriptions.
- European Parliament — Artificial Intelligence Act (AI Act) (2024), Official Journal of the EU.
- European Parliament — General Data Protection Regulation (GDPR) — Consolidated text, with subsequent amendments on automated data processing.
- Cerf, Vint & Kahn, Bob — “A Protocol for Packet Network Intercommunication” (1974), IEEE Transactions on Communications. Foundational work of the internet, for the historical context of networks.
- Microsoft Corporation — MS-DOS 6.22 User’s Guide (1994), archived at the Computer History Museum.
- OpenAI — “GPT-4 Technical Report” (2023), official documentation on the limitations and capabilities of language models.
- Zuboff, Shoshana — “The Age of Surveillance Capitalism” (2019), summaries and essays on the author’s website, for understanding the attention economy and digital dependency.
- Lessig, Lawrence — “Code and Other Laws of Cyberspace” (1999), online version at Harvard Law School.
- European Commission — “Ethics Guidelines for Trustworthy AI” (2019), High-Level Expert Group on AI.
- Petru Cojocaru — cojocarupetru.info, blog archive, articles on craftsmanship, technology, and economic transition (1997–2026).
Key Questions to Remember, with Eloquent Answers
1. Why was the 1997 green screen more honest than the 2026 AI interface?
Because the green screen hid nothing. It told you directly: “I don’t know what you want. Tell me exactly.” The 2026 AI interface smiles, appears friendly, but hides behind it data centres, hidden costs, and strategic dependencies. The blinking cursor was an honest challenge; the AI text box is an elegant trap.
2. What is the difference between a trainee who leaves after two weeks and an entrepreneur who buys an AI subscription without understanding it?
None. Both believe they have bought a solution when, in fact, they have bought a dependency. The 1997 trainee left with a certificate that did not make them an operator. The 2026 entrepreneur leaves with a €20 subscription that does not make them the master of the technology. Both confuse the paper with the craft.
3. Why did the 1997 entrepreneurs ask economic questions, not technical ones?
Because they did not come to become craftsmen of the tool, but to decide whether the tool was worth buying. The question “How much does it cost to keep records on a computer?” was more important than “How do you type DIR?” They delegated. But delegation requires a competent delegate, and competence comes at a cost.
4. What is “human verification of outputs,” and why is it a legal obligation in 2026?
It is the act by which a human, not a machine, assumes responsibility for a decision. AI can generate a formally perfect contract, but with a clause that ruins you. The law is clear: you cannot sue an algorithm. Responsibility remains with you. Human verification is the moment when the craftsman says: “This is not right.”
5. Why is local AI processing the digital equivalent of the 1997 AUTOEXEC.BAT file?
Because both represent the act of taking control of the environment in which the tool operates. In 1997, you modified AUTOEXEC.BAT to configure the machine to your needs. In 2026, you install a local AI model to avoid sending your data to the cloud. Both acts are declarations of independence.
6. What is the connection between the viruses of Windows 3.11 and the hallucinations of AI in 2026?
Both are intrusions of reality into the illusion of control. The 1997 virus showed you that the user-friendly interface hides vulnerabilities. The 2026 AI hallucination shows you that the simplicity of the interface hides a probabilistic engine, not a truth engine. Both teach you that ease of access comes with a price.
7. Why is legislation the “new command line” in 2026?
Because, just as in 1997 when you had to know the exact syntax of DOS commands, in 2026 you must know exactly what data you can process, where you can store it, and who is responsible for the output. Ignorance is no longer an excuse, but a fine. Legislation is the mandatory grammar of the digital age.
8. What does “digital sovereignty” mean, and why is it important for a small entrepreneur?
It means the ability to decide where your data resides, who processes it, and who has access to it. If your company’s data is on a server in California, you have outsourced not just a task, but a part of your business’s sovereignty. Digital sovereignty is the modern equivalent of the key to your office.
9. Why is the real cost of AI not the subscription, but the person who operates it?
Because AI, like MS-DOS, is mute in the face of those who do not know how to speak to it. An AI agent left unsupervised can spend a month’s budget in a day. You need someone to drive it, verify it, stop it when it goes off the rails. That person costs €2,500–€4,000 per month. The €20 subscription is just the entry fee.
10. What does it mean to be a “digital craftsman” in 2026?
It means not confusing the tool with the instrument. It means understanding that technology does not replace judgement, but demands sharper judgement. It means staying after the room empties, learning from mistakes, configuring the environment to your needs, and verifying every output with the critical eye of someone who has worked to understand. The digital craftsman is not the one with the most expensive subscription, but the one who knows what their certificate says.
Author’s Note: The Evolution of the Article’s Idea
This article was not born from a sudden inspiration, but from a real, mundane conversation that haunted me for weeks. In the autumn of 2025, I was invited to give a presentation about Artificial Intelligence at a meeting of local entrepreneurs. The room was full. People came with the hope of learning how to replace an employee with an algorithm, how to automate everything, how to escape the burden of thinking.
I began the presentation with a live demonstration. I generated text, created an image, analysed a data set. People were impressed. But then, during the coffee break, a logistics company owner asked me: “Well, and who operates it? Because I don’t have time.”
The question struck me. Not because it was new, but because it was old. I had heard it before. In 1997, in the same training room, another owner, with different clothes and different problems, had asked the same thing about MS-DOS: “Who operates it? How much does it cost?”
That was when I understood that I was not talking about technology. I was talking about dependency. About the illusion that a new tool frees you from work, when in fact it makes you dependent on the person who knows how to use it. I started writing the article that evening, but I did not yet know where it was going.
Initially, I wanted to write only about the parallel between 1997 and 2026. But as I wrote, I realised that the parallel was too simplistic. It was not just about “then vs. now”. It was about the structure of the relationship between humans and tools—a structure that repeats itself, that does not change, regardless of the technology.
I then added the section about Windows 3.11 because that was where I first saw how simplicity becomes a trap. The graphical interface seemed like a liberation, but it hid complexity and introduced new vulnerabilities. Similarly, the AI of 2026 seems like a liberation, but it hides dependency on infrastructures you do not control.
The legislation section came last because I realised that you cannot talk about technology in 2026 without talking about rules. In 1997, legislation in the digital field was almost non-existent. In 2026, it is omnipresent. GDPR, AI Act, data sovereignty—all of these are “command lines” that you must know, or you will get an error.
This article is not a nostalgia for a bygone era. It is a map for navigating the present. It is an attempt to show that, although the screens have changed, the psychology of the person sitting in front of the unknown has remained the same. Fear, curiosity, dependency, the illusion of control—all are the same.
I wrote this article for those who stay after the room empties. For those who are not satisfied with the certificate, but want to understand what it says. For the digital craftsmen who know that true competence is not measured in paid subscriptions, but in the hours spent in front of the screen, learning from mistakes.
Critical Analysis: Limits, Tensions, Comparisons, Subjective Errors
Limits of the Article
This article has a fundamental limit: it is subjective. The story is told from the perspective of a craftsman who has experienced both eras, who stayed after class, who learned from mistakes. This perspective excludes the experiences of those who did not stay, who left after two weeks, who considered that technology was not for them. This article is not an objective history of computing in Romania, but a personal meditation on the relationship between humans and tools.
Another limit is generalisation. Not all entrepreneurs in 1997 asked economic questions. Some came with a genuine passion for technology. Not all entrepreneurs in 2026 are disappointed by AI. Some have found real solutions, have automated processes, have increased efficiency. The article selects a certain typology of entrepreneur—the pragmatic one, the one who calculates costs—and ignores other typologies.
Narrative Tensions
The article is built on a fundamental tension: between apparent simplicity and hidden complexity. This tension manifests itself on several levels:
- Between interface and infrastructure: The 1997 green screen seemed simple, but hid the complexity of the operating system. The 2026 AI interface seems simple, but hides the complexity of data centres, algorithms, and costs.
- Between certificate and competence: The paper that certifies a competence is not the competence itself. The “IBM PC Operator” certificate from 1997 and the “Prompt Engineer” certificate from 2026 are both bureaucratic proofs, not proofs of craftsmanship.
- Between tool and instrument: A tool is something you use. An instrument is something you master. The distance between them is built by you, not by the manufacturer.
Comparisons That Work
- MS-DOS and AI: Both are “mute” to those who do not know how to speak to them. Both require precision in formulating the request. Both can generate spectacular results or catastrophic errors, depending on the user’s competence.
- Windows 3.11 viruses and AI hallucinations: Both are intrusions of reality into the illusion of control. Both teach you that ease of access comes with a price.
- AUTOEXEC.BAT file and local AI processing: Both represent the act of taking control of the environment. Both are declarations of technological independence.
Comparisons That Do Not Work (Incorrect Perspectives)
- Forced parallel between 1997 and 2026: Although the structure of the relationship between humans and tools is similar, the economic and technological context is radically different. In 1997, Romania was in transition, the economy was fragmented, and technology was accessible only to those who could afford it. In 2026, Romania is integrated into the EU, the economy is globalised, and technology is (theoretically) accessible to everyone. The parallel risks ignoring these fundamental differences.
- Romanticising the MS-DOS era: The article risks painting the ’90s as an era of digital innocence, when humans were in control. But the reality was harsher: most people did not have access to computers; those who did were dependent on instructors, specialist magazines, and limited communities. It was not an era of independence, but of dependency on a precarious information infrastructure.
- Demonising AI: The article does not demonise AI, but it risks being interpreted as such. AI is not “evil”. It is a powerful tool that can generate real value. But, like any powerful tool, it requires competence to be used correctly. The problem is not the technology, but the illusion that technology frees you from work.
Possible Subjective Errors
- Selective memory: The article selects certain memories from 1997 and ignores others. For example, it does not mention the real frustrations of working with MS-DOS: lack of support, inexplicable errors, data loss due to software bugs. These frustrations are softened in favour of a narrative about “learning through mistakes.”
- Generalising personal experience: The article presents personal experience as a universal experience. But not everyone who learned MS-DOS in 1997 had the same experience. Some had access to better computers, better instructors, more supportive communities. The classroom experience was not representative of all of Romania.
- Ignoring the global context: The article focuses on local experience but ignores the global context of technological development. MS-DOS was not invented in Romania; AI is not developed in Romania. Romania is a consumer of technology, not a producer. This structural dependency on imported technology is not sufficiently explored in the article.
“Lizards Explained” — Aesopian Language
The article uses Aesopian language, meaning it says one thing but implies another. Here are the “lizards” in the text, explained:
1. “The Green Screen”
Surface: The MS-DOS computer screen, with no instructions.
Depth: The state of ignorance, of the unknown, of lack of direction. The green screen is a metaphor for the moment when you face a new technology and do not know where to start. It is also a metaphor for the lack of transparency of technological systems: behind the user-friendly interface, there is a hidden complexity, a “green screen” that you do not see.
2. “The Blinking Cursor”
Surface: The visual indicator showing you where to type.
Depth: Challenge, expectation, judgement. The blinking cursor is a metaphor for the pressure to perform, to not make mistakes. It is also a metaphor for the passing of time: each blink is a second in which you have done nothing, a second in which the machine waits, judges, patiently.
3. “The Instructor’s Floppy Disk”
Surface: The physical storage medium, a small black plastic box.
Depth: Fragility, dependency, risk. The floppy disk is a metaphor for the vulnerability of data, for the fact that everything can be lost due to a single wrong move. It is also a metaphor for the centralisation of control: the instructor holds the floppy disk, therefore holds the knowledge, therefore holds the power.
4. “The IBM PC Operator Certificate”
Surface: An official document certifying a competence.
Depth: The illusion of competence, the confusion between paper and craftsmanship. The certificate is a metaphor for all certificates, diplomas, and accreditations that do not guarantee real competence. It is also a metaphor for bureaucracy: the system demands formal proofs, not proofs of ability.
5. “The €20 Subscription”
Surface: The monthly cost of an AI service.
Depth: The trap, the illusion of simplicity, hidden dependency. The subscription is a metaphor for all “ease” promises that hide real costs. It is also a metaphor for the externalisation of control: you pay €20, but you become dependent on an infrastructure you do not control.
6. “The €3,000 Project Manager”
Surface: The monthly cost of a specialist who knows how to operate AI.
Depth: The real cost of technology, the price of dependency, the illusion of automation. The project manager is a metaphor for the fact that technology does not replace humans, but demands a more competent human. It is also a metaphor for the hierarchy of competence: not all people are equal in the face of technology; some are more valuable because they know how to handle it.
7. “Viruses”
Surface: Malware programs that corrupt the system.
Depth: Vulnerability, intrusion, the illusion of control. The virus is a metaphor for all the hidden risks behind simplicity. It is also a metaphor for dependency on external sources: every borrowed floppy disk, every downloaded file is a potential threat.
8. “AUTOEXEC.BAT”
Surface: MS-DOS configuration file.
Depth: Control, independence, craftsmanship. AUTOEXEC.BAT is a metaphor for the act of taking control of the environment, of configuring the tool to your needs. It is also a metaphor for deep knowledge: the one who knows how to modify AUTOEXEC.BAT is not just a user, but a configurator, a craftsman.
9. “Digital Sovereignty”
Surface: The legal concept regarding control over data.
Depth: Independence, autonomy, resistance. Digital sovereignty is a metaphor for the ability to decide where your data resides, who processes it, who has access to it. It is also a metaphor for resistance to centralisation: local AI processing is an act of resistance against tech giants that control the global infrastructure.
10. “The Digital Craftsman”
Surface: The person who knows how to use technology.
Depth: Competence, patience, judgement. The digital craftsman is a metaphor for the one who is not satisfied with apparent simplicity, but seeks to understand the hidden complexity. It is also a metaphor for responsibility: the craftsman is not the one with the most expensive subscription, but the one who verifies every output, the one who assumes responsibility for the decisions made based on technology.
Rhetorical Devices Used in the Article
1. Anaphora (Repetition at the Beginning of Sentences)
Example:“It is not a magic button. It is not simple. It is an interface...”Effect: Creates rhythm, emphasises the contrast between expectation and reality.
2. Extended Metaphor
Example:“The computer is a capricious employee.”Effect: Transforms an abstract concept (technology) into a concrete, easily understandable image.
3. Structural Parallelism
Example:“In 1997, the question was... In 2026, the question is...”Effect: Creates symmetry, underlines the continuity between the two eras.
4. Irony
Example:“The certificate now has a different name. It no longer says ‘IBM PC Operator’. Now it says ‘Prompt Engineer.’”Effect: Highlights the absurdity of bureaucracy, the confusion between paper and competence.
5. Antithesis (Contrast)
Example:“Apparent simplicity vs. hidden complexity.”Effect: Creates narrative tension, underlines the fundamental tension of the article.
6. Enumeration
Example:“Text generation, image creation, complex data analysis, automation, autonomous agents...”Effect: Creates an impression of abundance, of infinite possibilities, which is then contradicted by the reality of costs.
7. Reticence (Unfinished Sentence)
Example:“A brief silence in the room...”Effect: Creates suspense, invites the reader to mentally complete what is not said.
8. Rhetorical Question
Example:“Do you know what it says?”Effect: Challenges the reader, involves them directly in the narrative, forces them to reflect.
9. Personification
Example:“The computer does not get angry. It does not get tired. It does not judge.”Effect: Humanises the machine, creates an emotional relationship between humans and the tool.
10. Epiphonema (Generalising Conclusion)
Example:“Technology does not replace the craftsman—it demands a better craftsman.”Effect: Summarises the article’s message in a memorable sentence that can be quoted independently.
Call to Action: What You Can Do Tomorrow?
5 Concrete Suggestions
1. Ask Yourself: “Who Operates It?”
Tomorrow, when you face a new technology, new software, or a new service, ask yourself: “Who will operate it? Who will verify the outputs?” Do not buy the subscription before answering this question. The real cost is not the licence, but the person who uses it.
2. Modify Your AUTOEXEC.BAT
Tomorrow, take control of a tool you use. Do not settle for default settings. Configure the environment to your needs. Whether it is productivity software, a workflow, or a business process, personalise it. Do not be just a user—be a configurator.
3. Verify an AI Output
Tomorrow, when you use AI to generate text, analysis, or a decision, verify it. Do not accept it blindly. Read it with a critical eye. Ask yourself: “Is it correct? Is it complete? Is it relevant?” Do not confuse interface simplicity with problem simplicity. Human verification is the supreme act of the digital craftsman.
4. Ask Yourself: “Where Is My Data?”
Tomorrow, when you use a cloud service, ask yourself: “Where is my data? Who has access to it? What happens if the service shuts down?” Do not externalise digital sovereignty without understanding the costs. Your data is your merchandise. Do you know where your merchandise is?
5. Stay After Class
Tomorrow, when you attend a presentation, course, or conference, stay after the room empties. Do not settle for surface information. Seek to understand the hidden complexity. Talk to those who know more than you. Learn from their mistakes. Craftsmanship is built in the hours no one has scheduled.
What to Avoid: Actions Not to Take
1. Do Not Buy the Subscription Before Understanding the Real Cost
Do not be fooled by the low entry price. The €20 subscription is just the beginning. The real cost is the person who operates it, the infrastructure that supports it, the dependency it creates. Calculate the total cost, not just the entry cost.
2. Do Not Confuse the Certificate with Competence
Do not settle for a certificate, diploma, or accreditation. These are bureaucratic proofs, not proofs of ability. Competence is measured in hours spent in front of the screen, in repaired mistakes, in verified outputs. Paper does not make you a craftsman.
3. Do Not Accept AI Output Without Verification
Do not confuse interface simplicity with problem simplicity. AI is a probabilistic engine, not a truth engine. It can generate hallucinations, make mistakes, violate legislation. Human verification is not optional—it is mandatory. You cannot sue an algorithm.
4. Do Not Externalise Digital Sovereignty Without Understanding the Costs
Do not send your data to the cloud without knowing where it is, who has access to it, what happens if the service shuts down. Digital sovereignty is the modern equivalent of the key to your office. Do not externalise it without understanding the costs.
5. Do Not Settle for Apparent Simplicity
Do not confuse a user-friendly interface with real simplicity. Behind every user-friendly interface, there is hidden complexity, infrastructure, dependency. Seek to understand the complexity. Do not be just a consumer—be a craftsman.
6. Do Not Ignore Legislation
Do not think legislation is just bureaucracy. Legislation is the “new command line”. Ignorance is no longer an excuse—it is a fine. Learn the rules, respect them, apply them. Legislation is not an enemy of innovation, but the guardian that forces you not to forget who is actually in charge.
7. Do Not Romanticise the Past
Do not believe the MS-DOS era was a golden age of digital innocence. It was an era of dependency on precarious information infrastructure, lack of support, inexplicable errors. Do not confuse nostalgia with reality.
8. Do Not Demonise the Future
Do not think AI is “evil”. It is a powerful tool that can generate real value. But, like any powerful tool, it requires competence to be used correctly. The problem is not the technology, but the illusion that technology frees you from work.
9. Do Not Confuse the Tool with the Instrument
A tool is something you use. An instrument is something you master. The distance between them is built by you, not by the manufacturer. Do not buy the most expensive tools if you are not prepared to turn them into instruments.
10. Do Not Leave Before the End of the Course
Do not leave the classroom, presentation, or conference before truly understanding. Do not settle for surface information. Seek to understand the hidden complexity. Craftsmanship is built in the hours no one has scheduled. Stay after class.
Petru Cojocaru is the founder of ABSOLUT WEB EXPERT SRL, a web agency based in Botoșani. He writes at cojocarupetru.info.
AI Declaration: This article was written with the assistance of a language model (Qwen) for structuring and bibliography. The editorial voice, personal experiences, and value judgments belong entirely to the author.