❯AI psychosis and technological realism

For the things we have to learn before we can do them, we learn by doing them.

—Aristotele

Looking back at the last decade I have noticed a behavioral pattern that is common to experienced people. It's not so much the use of advanced concepts, though that does play a role, but it's their thought patterns oriented towards of risk avoidance, complexity control and maintainability. These patterns are acquired via years of work and they manifest themselves as a sort of intuition. I call that intuition because typically you don't "study risk", you learn what it means by making mistakes, and it gets imprinted into your mind with each painful lesson.

Effectively engaging with complex systems, whether those we’ve built through technology or those shaped by history’s painful progression, requires steady maintenance, error avoidance and looking at the big picture, rather than chasing novelty. Though new solutions may offer real advantages, reality ultimately favors preservation and cautious updates over reinvetion.

AI, by contrast, lacks this experiential anchor. It operates in a vacuum of "average" probability, following trends rather than sound engineering principles while claiming the novelty, and the efficacy, of the proposed "solutions". While a human expert knows when to backtrack and pick a different approach, AI agents lack moderation and strategy, and their solutions don't seem to be built on solid foundations by careful selection. Practically speaking many developers resort to maintaining an array of Markdown files and constantly prompt AI to fix its broken code to get close to the quality they used to deliver.

It has become difficult to approach this topic meaningfully. There’s the emergence of a type of person who outright dismisses any concern about AI and replies with superficial rebuttals that avoid addressing the core issue. I'm not the only one who has noticed this.

There is a common trend of pointing at flashy, AI-generated code and saying, 'Look, it works!' as if that settles the debate. But that’s a dangerous fallacy. Even if a specific function compiles perfectly, the wider project could easily be corrupting data, triggering silent errors, or completely collapsing under its own weight.

Other people have noted that models like Claude often translate every problem into "more code" rather than "less code". Here's a brief synopsys:

• AI returns 1,000 lines of "crazy non-performant crap" for a task that an expert could implement in ten times fewer lines of code, and more elegantly.
• You can't possibly read, review and understand multiple tens of thousands lines of code per day, nor per week. Especially if we're talking about complex languages and projects.
• By offloading tasks, not only the repetitive ones, you stop exercising useful skills, your performance and understanding degrades. Hardly any skill stays sharp if it isn’t practiced for an extended period.
• "Vibe coding" just means endlessly setting up and nurturing configurations to allow AI models that you do not control to do the job, with all mentioned risks, instead of directly doing the job right.

Conclusion

Ultimately, engineering is a thinking profession, not a typing one. The real value of an expert developer lies in making the right choices at the right time, choices that can save a company months or years of rework. Until enterprises understand risk, they will continue to pay a hidden, significant price for the "speed" of AI.