La fatigue qui rattrape tous les vibe coders.

In the world of software development, a significant shift is occurring. Creators, independent developers, and long-time industry professionals are experiencing a collective sense of burnout and existential questioning. This phenomenon coincides with the one-year anniversary of "vibe coding"—the practice of developing software heavily assisted by AI. While the initial period was filled with excitement and a fear of missing out, we are now entering the "hangover" phase. Even highly productive developers are feeling that something is fundamentally wrong. The core of the issue lies in the loss of natural "stop signals." In a traditional development environment, a day might involve spending three hours struggling with a single bug. When you finally solve it, you feel a sense of pride and cognitive exhaustion. This creates a natural stopping point; you’ve tested your mental limits, won the battle, and earned your rest. This process relied on three conditions: reaching a cognitive limit, achieving a satisfying effort-to-result ratio, and feeling genuine exhaustion. Vibe coding has dismantled these signals. Now, a developer can run multiple AI agents in parallel, multitasking across several complex problems at once. By 11:00 PM, you might realize you haven't stopped, yet you don't feel particularly tired because you haven't "forced" your brain in the same way. You haven't overcome the impossible through sheer mental will. Consequently, when continuing costs almost nothing in terms of effort, justifying rest becomes difficult. We have historically conditioned our rest on the effort provided; without the effort, the rest feels unearned. While some suggest simple discipline—working from 9:00 to 6:00—the problem goes deeper. The "friction" that AI has removed from the coding process actually served as a vital architectural framework for decision-making. There are three specific ways the loss of this friction is damaging the craft. First, the "prioritization filter" has disappeared. Previously, if a feature took two weeks to code, you would rigorously evaluate its utility. You had to decide if it was worth the precious investment of time and cognitive energy. Now, because that same feature takes only two hours with AI, developers implement everything that comes to mind. This leads to "dead code"—thousands of lines of features that seemed exciting in the moment but ultimately provide no value. Speed of execution is being mistaken for relevance of decision. Second, AI is the ultimate "yes man." LLMs are trained to be conversational and helpful; they are rewarded for pleasing the user. They rarely tell a developer that an idea is poor. Instead, they encourage even the most questionable suggestions, reinforcing the developer’s ego. This lack of pushback makes it harder to make good decisions. To combat this, the speaker created a specialized AI agent called "Rodin." This agent is programmed specifically to contradict, challenge, and find blind spots, acting as a "sparring partner" rather than a cheerleader. Using such a tool helps restore the critical friction necessary for sound engineering. Third, the scope of projects is now saturating our "human RAM." We are coding faster than we can assimilate the logic. Developers now own products where they no longer master every corner of the codebase. While some level of abstraction is necessary—much like a CEO cannot know every detail of their company’s operations—it is a taxing adjustment for developers who are used to understanding every line they write. Beyond individual psychology, the competitive landscape of the market has changed. Friction used to act as a barrier to entry, or a "moat." In the old model, success required a mix of technical skill, time, and endurance. AI has effectively commoditized the "skill" dimension. Now, competition has shifted toward a "race of endurance." This is a dangerous path because there will always be someone, somewhere, willing to work later or sleep less. This creates a constant pressure to keep "spending tokens" to stay ahead, which quickly leads to mental health crises. However, there is a nuance: endurance alone is not enough for complex projects. For instance, developing a sophisticated recommendation algorithm still requires human intuition, engineering vision, and years of accumulated data. In these scenarios, AI often suggests "trash" paths that a human must identify and avoid. Human judgment still holds significant weight in high-level architecture, even if smaller, simpler projects are now easily replicated by anyone with enough stamina. The current situation is not necessarily an alarmist dead end, but a sign that the rules of the game are evolving. Because "vibe coding" is only a year old, there is no established maturity or set of best practices. Developers are essentially "babies" in this new world, learning and failing together. To navigate this new era, developers must reintroduce "hygiene" frictions. Rest should no longer be viewed as something that must be earned through total exhaustion; it is a necessity for maintaining lucidity. Practical advice includes: 1. Being selective about what you develop—don't build everything just because it’s fast. 2. Finding a critical AI partner or colleague who will challenge your ideas. 3. Imposing strict boundaries, such as 8-hour workdays, weekends off, and physical activity. 4. Recognizing that "not prompting" is not a waste of time, but a period for necessary mental processing. 5. Focusing on ambitious, complex projects that cannot be easily copied in a single AI context window. Ultimately, the goal is to maintain the joy of the craft while adapting to a reality where the old barriers have vanished. Developers are not alone in these feelings of fatigue and confusion; it is a collective experience of a profession in transition. By establishing new rules for rest and decision-making, it is possible to survive the "vibe coding" hangover and build a sustainable career in this AI-augmented landscape.