It had been some time since I last wrote a reflection like this. Over the past few months, however, I have paid closer attention to the fear many people have developed around artificial intelligence.
Some refuse to use AI tools because they do not know what will happen to their data. The concern is legitimate. What is curious is that some of those same people publish details of their lives on social media, grant excessive permissions to apps, and accept terms of service without knowing what information is being collected.
Artificial intelligence did not create the problem of digital privacy.
Long before generative AI became popular, search engines, social networks, mobile applications, devices, and advertising platforms were already collecting data that could reveal habits, interests, movements, and relationships. Google itself provides controls to review, delete, or prevent the storage of activities such as searches, browsing history, watched videos, and location.
AI entered a world that had already turned personal data into an economic asset. It expanded our ability to process that information, but it did not create the model.
The fear of artificial intelligence did not begin with AI
Artificial intelligence arrived much like the internet did: creating distrust in some people, excessive enthusiasm in others, and a commercial race among companies.
The difference is speed.
In software engineering, we can study a tool today, test it tomorrow, and discover the following week that another solution already performs the same task differently, or better.
This pace makes the moment fascinating and uncomfortable at the same time. Nobody knows precisely how far these technologies will go. Discussing risk, privacy, security, transparency, and accountability is therefore not alarmism; it is necessary.
The problem begins when technical debate is replaced by sensationalism.
Some people have realized that fear generates attention, reach, and headlines. Extreme predictions are then presented as certainties to a public still trying to understand what artificial intelligence is and what it can actually do.
The risk is real, but it does not arise on its own
The title of this article is intentional.
For thousands of years, humans have waged wars, invaded territories, enslaved, manipulated, robbed, and killed other humans. Yet we often speak about artificial intelligence as though it were the first major threat created by our species.
AI systems clearly present real risks. They can reproduce bias, expose information, generate false content, support flawed decisions, or carry out actions with unexpected consequences.
The NIST AI Risk Management Framework, one of the leading international references on the subject, treats safety, privacy, reliability, transparency, and accountability as essential parts of responsible AI development.
I am not arguing for unrestricted trust in machines.
What I question is how easily we turn technology into the sole villain while ignoring who defines its objectives, selects its data, grants its permissions, and decides how its output will be used.
In films where machines dominate humanity, there is usually an earlier human decision: someone pursued control, advantage, efficiency, or power without properly considering the consequences.
Before fearing the machine, I still fear the ambition of those who control it.
While we debate limits for artificial intelligence, people continue using far less sophisticated technologies to surveil, manipulate, exploit, and destroy. There is a contradiction in that.
What changed in software engineering
A few months ago, I structured an experimental software engineering squad made up of specialized AI agents.
From initial requirement analysis through documentation, development, code and architecture review, testing, security validation, and deployment preparation, different agents participated in the process. My role shifted toward coordinating the work, defining criteria, reviewing decisions, and authorizing every meaningful step.
Soon afterward, tools such as Cursor and Claude Code began incorporating similar capabilities into their own environments. Claude Code, for example, supports specialized subagents, each with its own context, instructions, tools, and permissions.
My agents did not lose their usefulness. They still follow the way I prefer to structure and validate development. What changed was the market: capabilities that once required a custom architecture began appearing as native features in commercial tools.
As I often say, including in conversations at work, I do not see a better way to separate what is real from what is not than by using these tools and experimenting with them. We will need to test and build things with AI. Otherwise, we remain far too exposed to every sensational headline that comes along.
As software engineering professionals, we cannot be afraid to use it. We can limit the impact: build a small product, write a script, or run a controlled experiment. I choose the model, decide whether to use ChatGPT, Gemini, Grok, or another tool and, most importantly, control how much autonomy I am prepared to give it.
This is exactly what I mean by learning through building. I learn better in practice. Without skin in the game, we become much more likely to believe everything we are told, both the exaggerated promises and the catastrophic predictions.
This is a portrait of today's speed.
Learning a tool is no longer enough. We must learn to study, experiment, and adapt continuously.
Professions will not remain as they are
Saying that no profession will disappear would be as premature as claiming that AI will eliminate every professional.
Some tasks will be automated. Certain roles may lose demand, while others will be created or deeply changed. The most likely outcome, however, is that many occupations will be transformed rather than simply eliminated. That is also the central conclusion of the global study published by the International Labour Organization.
Software engineering will increasingly need professionals able to:
- break down complex problems;
- define context and acceptance criteria;
- choose the appropriate model or agent;
- set autonomy boundaries;
- review architecture, code, security, and impact;
- identify plausible but incorrect answers;
- take responsibility for the delivered result.
The engineer of the future may look less like someone manually operating every tool and more like the person responsible for a large automated workshop.
They do not need to execute every movement personally, but they must know the machines, understand their limits, and know when, or when not, to activate them.
This does not reduce the importance of technical knowledge. On the contrary: the greater AI's execution capacity, the greater the responsibility of the person coordinating it.
Autonomy without governance remains reckless
I do not support granting unrestricted autonomy to artificial intelligence. But I also do not reject autonomy merely because a machine is carrying it out.
An ambiguous instruction can lead a human team to produce the wrong result. The same instruction can make a team of agents fail at greater speed and scale.
The difference lies in limiting the scope of action, recording decisions, restricting tools, requiring validation, and placing human approval at critical points.
Machines can consult large volumes of rules and execute repetitive processes with a consistency that is difficult to maintain manually. That does not mean they have perfect memory or follow instructions without failure. Models have context limits, can misinterpret rules, and can produce convincing but false answers.
The human professional therefore does not disappear from the process. Their role shifts from executing every step to designing, supervising, and being accountable for the system.
Should we be afraid of artificial intelligence?
Do not reject AI simply out of fear.
Learn how it works. Understand its limits. Define what data it may use. Restrict its access. Validate its answers. Maintain oversight proportional to the risk, and never delegate to a machine a responsibility that remains yours.
At the same time, do not assume that every concern comes from misinformation. There are technical, social, economic, and political risks that deserve serious discussion.
But we must frame the question correctly.
Artificial intelligence does not independently possess corporate ambition, a political project, or a desire for domination. It amplifies the capacity of people and organizations, including those who already held power before it existed.
Perhaps the central question is not:
“What will artificial intelligence do to us?”
Perhaps we should ask:
“What will humans decide to do with the power artificial intelligence has placed in their hands?”
Our future will depend on that answer.