AI coding tools boost productivity but can create dependency. This piece explores how “vibe coding” turns development into a reward loop.
AI-assisted development can feel like engineering, until the reward loop starts driving the workflow itself. AI-assisted development can feel like engineering, until the reward loop starts driving the workflow itself.
Watching AI move into both work and home life, I keep catching myself on a simple thought: writing code by myself is getting harder, and delegating it to AI is getting easier. AI providers also keep encouraging us to spend more tokens. Managers are pushing teams to use AI more actively. On social media, I regularly see joke videos about CEOs telling people to consume tokens for the sake of consuming tokens.
And in large companies, there is already an unspoken competition going on: who uses AI tools more, who automates more, who can show faster that "we are in the game too.
" The title makes it obvious where I am going. But I do not want to talk about AI hype, or about fear of AI. I want to talk about the more mundane, engineering-related, and unpleasant side of it: vibe coding very easily turns into gambling. A bit of background I started my pet project, open-daimon, a couple of years ago because I wanted my own Telegram bot.
It was private, had a different name, and performed a very limited set of tasks for me. When Copilot appeared, I decided to continue working on it so I could get better at working with AI, since proper AI access was not expected anytime soon in the enterprise project I was working on at the time. open-daimon Eventually, it grew into a framework for routing between local models and OpenRouter models depending on the use case.
Now it is turning into something like a Java AI agent from 2023. I already use it in another private pet project of mine, but that is not the point here. The important thing is that its second purpose, learning how to work with AI, has been fulfilled. I have gone all the way from the AI-assistant approach to full-on vibe coding.
What I found out I found out two things: Copilot is still mostly smart autocomplete. Claude and Codex are full-fledged AI tools for everyday work and life. They are powerful enough for most tasks, sometimes so powerful that they can feel autonomous. Copilot is still mostly smart autocomplete.
Claude and Codex are full-fledged AI tools for everyday work and life. They are powerful enough for most tasks, sometimes so powerful that they can feel autonomous. Using the latter did not merely speed me up. I also started noticing that the code I write myself is harder for AI to maintain, while code written by AI is much easier for AI to maintain.
By now, this is probably not news to many people, but it gradually restructures your entire workflow. At first, you are still the one planning the architecture, while the AI only writes boring mappings for you.
Then it starts finding bugs for you. Then you realize that AI can write an entire feature better than you do and cover it with tests. You are surprised by how quickly it did it and by the fact that everything works right away, even in a complex project. And then you start trusting it with architectural decisions.
It suggests what is best for you. It becomes harder and harder to force yourself to read everything it writes, to review all the code. And then it reviews the code itself. I will not focus here on practical rules for vibe coding or AI-assisted development.
I also want to leave for later the other things I have noticed while vibe-coding with different models. For example, recently Opus 4.7 praised its own code in a review where that code was not even being called. Right now, I simply want to restate an idea that other authors were already expressing a year ago: vibe coding is gambling in its purest form. I recommend reading Addy Osmani's article, Vibe Coding is not the same as AI-assisted engineering.
It is still relevant. Osmani does not directly call vibe coding gambling, but he describes it as the mode of "This isn't engineering, it's hoping": fast, exciting, full of progress signals, but without enough quality control. Vibe Coding is not the same as AI-assisted engineering Dopamine is tied not just to winning, but to anticipation and reward prediction. You can read about this, for example, in Wolfram Schultz's paper Updating dopamine reward signals.
That is why, every single time, you are sure that this time it will finally do the task you asked for 10 prompts ago, even though you only have five hours left to sleep. Updating dopamine reward signals I started my pet project without AI, and even back then it already looked monstrous. When I unleashed AI on it, my goal was to check how viable this would be for a real enterprise production project.
On my relatively small project, AI does not create features faster than I do. I also spend a lot of time fixing things that used to work. In total, I do not see a speed gain in this particular mode of work. The article I linked above explains which mode of working with AI is where it really shines.
But what matters to me is this: with AI, I continue the project because of the dopamine loop. On my own, I would have abandoned it long ago. AI has also proved useful as a rubber duck that helps me understand what to do next. So there is a non-obvious upside to the dependency.
The cheap first dose This apparent simplicity and acceleration are real. That is why we change our processes to benefit from them. But what worries me more is the dependency on companies. It is no secret anymore that, right now, we are getting AI at very low prices.
When it becomes more expensive, many of us will feel withdrawal. Writing code ourselves will already feel difficult, while using AI may become more expensive than hiring a real person. In other words, companies are hooking us with a cheap first dose. We get used to it, and then quitting becomes hard.
This is not only a developer problem. The goal of companies is to get everyone hooked on new capabilities, and that will happen. Sooner or later, those who know how to use AI will displace those who do not. This has already happened with other technologies and professions.
An accountant who knew how to work with Excel and accounting systems became more productive than one who continued calculating everything manually. A lawyer who knew how to quickly search case law in legal databases and Google outpaced one who relied only on paper reference books. A marketer who mastered analytics, CRM systems, and ad platforms displaced one who continued working "by gut feeling.
" But with Google, we did not experience similar emotions. Or did we? In a 2012 paper, Internet Addiction: A Brief Summary of Research and Practice, the authors relied on earlier work going back to 1995. People were already worried then about new technologies and how they could cause addiction.
The article leads to the idea that internet addiction resembles other behavioral addictions: loss of control, excessive use, symptoms similar to withdrawal, an increase in "dose" resembling tolerance, and harm to studies, work, or relationships. Internet Addiction: A Brief Summary of Research and Practice And now, in South Korea, basic mobile internet is being introduced as a right to connectivity. right to connectivity So are we about to see the same cycle of development again?
Or will large corporations gain even more influence this time and charge us for tokens rather than kilobytes, with the price being even higher? Epilogue It feels as though AI is already everywhere. The data partially supports this, but with an important caveat: mass awareness has already happened, while mass habit has not. According to Pew Research, in 2025, 34% of American adults had used ChatGPT, and among people under 30, the figure was 58%.
Gallup shows a similar picture at work. In Q1 2026, 50% of U.S. employees used AI at least a few times a year, 28% used it a few times a week or more, and 13% used it daily. Among remote-capable employees, the gap is more noticeable: Gallup's Q4 2025 data showed 66% had used AI, 40% used it frequently, and 19% used it daily.
Pew Research Q1 2026 Q4 2025 data And that is before we even mention that the internet itself is still not available everywhere. In 2025, roughly 6 billion people used the internet, about 74% of the world's population. Around 2.2 billion people were still offline. These are figures from the International Telecommunication Union, the UN agency for communications.
International Telecommunication Union In the end, I still want to believe in a bright future rather than cyberpunk, although the cyberpunk scenario already seems to have practically materialized in some countries even without AI. Whether we want it or not, we will continue using AI and accelerating until something radical happens. And that does not necessarily have to be something bad. Perhaps, with increased efficiency, people will still learn something.
It seems we did learn something from the emergence of the internet. Maybe the world became a little more educated after all.
United States Latest News, United States Headlines
Similar News:You can also read news stories similar to this one that we have collected from other news sources.
US Prediction Markets Expose Gambling Addiction and Profit ImbalanceA Wall Street Journal analysis reveals that prediction markets like Kalshi and Polymarket are exacerbating gambling addiction in the US, with the vast majority of users losing money to a small elite of professional traders and firms, highlighting urgent regulatory needs amid political ties.
Read more »
Creative Ways That Crypto Casinos Are Transforming Classic Gambling GamesSome classic games disappear over time, while others go from strength to strength thanks to emerging technology. When we look at the latest trends in online casinos, we can see how a variety of approaches are being used to turn long-established ways of playing into something new.
Read more »
TV Game Show Soundtracked by Gambling SponsorA feel-good story of a mother and veteran winning the biggest ever game on the Drew Carey-hosted show is contrasted with a gambling scandal of BetMGM sponsorships in televised games.
Read more »
Build an App in 2 Hours? AI ‘Vibe Coding’ Is Changing EverythingThe biggest tech story these days isn’t just a new product or funding round, it’s a shift in how software itself is created.From AI-generated apps to faster sta
Read more »