The Three Generations of Programmers — How Has Artificial Intelligence Affected Our Programming…
The Three Generations of Programmers — How Has Artificial Intelligence Affected Our Programming Abilities?
There is a sharp paradox lived by the programming community today. At a time when a developer with five years of experience completes a programming task in one hour a task that would have taken an experienced programmer a full week in the nineties the Stack Overflow 2025 survey, which covered more than 49,000 developers from 177 countries, indicates that 46% of developers do not trust the accuracy of AI tool outputs. Meaning, the tool that doubled production speed is not trusted by half of its users. How did we get here?
The answer is not in the tool itself, but in the journey that the programmer’s relationship with information has traveled. A journey of three distinct stations that redraw what it means to be a “programmer.”
Station One: Information Requires Physical Spatial Contact (Before 2000)
The internet was not widely spread during that period. The primary source of knowledge was large, tangible printed books. Examples of such books include the O’Reilly series collection or Microsoft Press publications, containing hundreds of pages, of which the programmer needed only a few paragraphs.
We are now in the year 2026, and I personally own a book dedicated to learning JavaScript titled The Complete Reference — JavaScript Third Edition by Thomas A. Powell and Fritz Schneider. You cannot imagine the size of the book! It is 954 full pages as a complete reference for the JavaScript language, and it was published back in 2012, that is, 14 years ago from today. The price of this book is $50, but I bought it a few years ago, before the spread and dominance of artificial intelligence, for a very low price, evidence of the impact of these developments on printed books. I will leave you with a photo of the book.
The search mechanism relied on the “keyword index” at the back of the book; you would sit and flip through pages until you reached the function or concept you were looking for. If the required book was not within reach, the alternative was to travel to a specialized library or wait for the book to arrive by mail.
The mental impact of this phase was profound, despite the slow learning process. The programmer who spent two hours searching for the cause of a specific error in a large book was building a complete map of the problem in their mind, understanding why the error occurred before understanding how to fix it. The cost of accessing information was high, but that cost was building depth and reinforcing the information.
Station Two: The Digital Alternative — Before Artificial Intelligence (2000–2022)
With the spread of the internet and the emergence of search engines, the equation changed completely. It was no longer about where the information existed, but about how you describe what you are looking for. The new skill was not in memorization, but in how to search and how to formulate a query.
Then Stack Overflow arrived in 2008 to create a particular shift a platform that triggered a true revolution in the way problems were solved, transforming the collective expertise of millions of programmers into an organized, voted-upon knowledge base. You no longer needed to reinvent the wheel; you only needed to know the right link that showed you how others had used it.
The numbers from that era’s rise tell the story clearly: the monthly question count on Stack Overflow reached more than 200,000 questions per month between 2014 and 2020. The platform was equivalent in value to an entire company until Prosus acquired it for $1.8 billion in 2021.
But there was another price for this evolution. A generation of “copy and paste” was born, and the skill no longer lay in understanding the code, but in the ability to filter: which answer is correct? Which is outdated? And which one fits your context exactly? The challenge shifted from “finding information” to “distinguishing between pieces of information.” This is an entirely different skill from building the deep mental models that the previous station used to forge. What I mean here is that during that period, you were copying the code and pasting it directly into your programming project, but you had to know whether this code suited your project’s structure or whether you would cause a programming error. Despite that, the skill was still built on understanding, enabling the ability to distinguish between the correct and incorrect solutions.
Station Three: Ask Artificial Intelligence (2023 — Today)
In November 2022, ChatGPT was launched. What followed was not merely a change in tools, but an earthquake that redrew the map of the skills required of a developer.
Instead of searching for “how to connect SQL Server to ASP.NET,” the programmer now writes: “Write me a C# class representing a Repository that handles the Users table while accounting for Dependency Injection.” The difference is fundamental: you are no longer searching for an article written by another human, but asking the machine to build a solution designed for your specific context.
Here are the latest numbers:
- More than 70% of developers use AI in their work in 2026.
- According to Stack Overflow, a previous statistic showed that 84% of developers use or plan to use AI tools in 2025, compared to 76% in 2024.
- 51% of professional developers rely on these tools daily.
- A study published by GitHub in collaboration with the arXiv journal proved that developers who used GitHub Copilot completed a programming task in 1 hour and 11 minutes, while their counterparts without the tool took 2 hours and 41 minutes that is, 55.8% faster. Not merely a feeling of speed, but a statistically verified result.
- Today, 41% of all code written is generated with the assistance of AI, according to 2025 reports.
But on the flip side, what has become of Stack Overflow, built at one billion eight hundred million dollars? It collapsed. The monthly question count dropped from 200,000 at its peak to fewer than 50,000 by the end of 2025, returning to its levels at launch in 2008. Fifteen years of community building were wiped out in less than two years. ChatGPT was trained on Stack Overflow data, then pulled its users away from it. The ironic part is that Stack Overflow is now selling that deteriorating content to the very same AI companies.
The Paradox That Puzzles Everyone
While the number of AI dependencies climbs, the curve descends in the exact opposite direction when it comes to trust. Positivity toward these tools declined from 70%+ in 2023 to 60% in 2025. And most critically: 46% of developers explicitly state a lack of trust in the accuracy of AI outputs, versus only 33% who do trust them. In fact, only 3% reach the level of “high trust.”
The Stack Overflow 2025 report reveals the deeper wound: 66% of developers said their biggest problem with AI is getting “solutions that look correct but aren’t quite right.” And more strikingly, 45% reported that debugging AI-generated code takes longer than expected. What was supposed to save time sometimes steals it instead.
Researchers at MIT found in their field experiments that the actual improvement in developer productivity at Microsoft and Accenture ranges between 7.5% and 21.8% — a real number, but far below the 55% announced in laboratory experiments. The gap between the controlled environment and actual reality says a great deal.
What Actually Changed Is the “Feedback Loop”
The true technical philosopher in this transformation is not the tool, but the speed of the loop. In the age of books, the feedback loop from problem to solution to understanding took days. In the age of Google and Stack Overflow, it has shrunk to minutes. In the age of AI, it became seconds.
The puzzling question here: does speed make information difficult to retain?
When a programmer spends two hours searching for the cause of an error, they build a mental model of the entire system in their mind. When they get the answer in ten seconds, they may obtain the solution without building that model. And perhaps this is why seasoned developers are the most skeptical of AI outputs, and not beginners. Those who built mental models know when the tool is lying.
Three Generations, Three Skills
What truly deserves reflection is that each phase did not actually replace the previous skill, but rather reordered its priorities:
In the age of books, the skill was patience and memory. And verifying code correctness was done through repeated experimentation.
In the age of Google and Stack Overflow, the skill became filtering and discernment. How to evaluate an answer depended on community votes, publication dates, and the context of the question.
In the age of AI, the new skill is prompt engineering and analytical criticism. Verification is no longer in the hands of the community, but in yours alone — and in the logical tests you prepare.
The programmer who does not possess a mental model will not know how to evaluate what AI produces. And this is precisely what the numbers reveal: the more the dependency grows, the more the doubt also grows, but only among those who possess the knowledge foundation to perceive the size of the gap.
My Personal View
To be frank, nothing in this world stays the same. Evolution is not a choice we embrace or reject, but a reality that imposes itself upon us. The real question is not “do we adapt?” but “how do we adapt intelligently?”
If you ask me about my position in the midst of this era, my answer is clear: do not stop practicing. Even if you belong to the generation that built its understanding on solid fundamentals the generation that spent hours reading documentation and tracing errors line by line the language of numbers confirms that the shift toward the new system has become inevitable. The danger does not lie in using AI, but in falling into the trap of dependency on it. Well-established skills are not immune forever; they are exactly like a muscle they weaken if neglected, no matter how much prior capital you have.
Your historical expertise is your true advantage today, but it remains an advantage only if you preserve it. The practical solution I see is maintaining problem-solving as a periodic practice, not as a memory. Dedicate regular time to solving programming challenges designed for your level, and use AI itself to suggest these challenges this is precisely the conscious use of the tool: do not ask it for the solution, but ask it to place you in a situation that requires you to think.
And alongside that, there is no escaping mastery of the skill of Prompting — that is, how to formulate your question with precision to obtain a result that is worth your time. The tool is not so different from any other tool in history: its value is determined by the capability of the one who holds it.
Conclusion
The journey of programming knowledge is not a story of simple linear progress. It is a story of continuous exchange between speed and depth. Each station gained in one dimension and lost in another:
- Books gave you depth and took away your time.
- Google and Stack Overflow gave you access and took from you memorization.
- Artificial intelligence gives you speed — but what does it take?
Perhaps the question is still open. But what is constant is that the programmer who can evaluate what AI produces not merely use it, is the one who makes the difference. And evaluation does not come from another tool, but from the understanding built with patience, resembling the patience of those who used to flip through the pages of heavy books.
Watch YouTube Video:
https://medium.com/media/42eae21105e2d19bbb11624bca823c59/href
Sources
- Stack Overflow Developer Survey 2025 — survey. stackoverflow.co/2025
- Peng et al. (2023), “The Impact of AI on Developer Productivity: Evidence from GitHub Copilot” — arXiv:2302.06590
- Stack Overflow Traffic Collapse Visualization — Sam Rose, January 2026
- VentureBeat: “Stack Overflow data reveals the hidden productivity tax of ‘almost right’ AI code” — December 2025
- MIT PubPub: “The Productivity Effects of Generative AI: Evidence from a Field Experiment with GitHub Copilot”
- Stack Overflow Press Release — “Stack Overflow’s 2025 Developer Survey Reveals Trust in AI at an All-Time Low”
The Three Generations of Programmers — How Has Artificial Intelligence Affected Our Programming… was originally published in Towards AI on Medium, where people are continuing the conversation by highlighting and responding to this story.
