Dear Diary: What Happens to Developers When They Code With AI Every Day

A groundbreaking RCT study followed developers for weeks as they worked with AI tools — revealing subtle emotional, cognitive, and creative shifts.

For all the talk about productivity, efficiency, and code quality, one question has rarely been asked: what does working with AI actually do to people? That’s exactly what the study “Dear Diary: A Randomized Controlled Trial of Generative AI Coding Tools” set out to answer. Over six weeks, researchers followed two groups of software developers — one using AI assistants daily (like GitHub Copilot, Cursor, and Lovable), and another working entirely without AI. Each participant kept a diary of their workday, recording emotions, frustrations, and reflections. The results paint a surprisingly human portrait of life inside the new AI-assisted workspace. The Emotional Curve At the beginning of the trial, developers in the AI group reported feelings of excitement and empowerment. Tasks that used to take hours were completed in minutes. Many described the experience as “addictive” — watching lines of code appear almost magically in real time. By week three, however, the emotional tone began to change. The same participants started reporting feelings of detachment, reduced satisfaction, and even creative flatness. One developer wrote in her diary: “I’m productive, but I don’t feel proud. It’s like someone else did my job — and I just approved it.” Another noted: “When I’m not using the AI, I feel slower. When I am, I feel… hollow.” The researchers called this phenomenon “creative displacement” — the subtle psychological shift when creative agency is outsourced to a machine. Productivity vs. Fulfillment Objectively, the AI-assisted group outperformed the control group across all measurable metrics. They wrote more code, solved more tickets, and reported lower stress levels related to deadlines. But when asked about fulfillment, their self-reported satisfaction dropped by 18% compared to the control group. In other words, AI helped them win the sprint — but drained some of the meaning from the work. Interestingly, the control group — who coded manually — showed higher frustration but also higher flow state engagement. Their work felt harder, but more personal. Trust and Dependency By week four, most AI-assisted participants had stopped questioning the AI’s output. Many described it as a “partner,” not a tool. This led to what the researchers termed “trust saturation.” Developers began skipping reviews or rationalizing obvious mistakes with lines like, “The AI probably knows better.” The same developers also reported anxiety when the AI was unavailable. Some admitted that they “froze” when forced to code manually again. It’s a subtle dependency that mirrors other forms of automation reliance — once humans offload decision-making, they rarely take it back. The Identity Question One of the most powerful themes in the diaries was the erosion of craft identity. Senior developers expressed concern that their expertise mattered less. Junior developers, meanwhile, felt “amplified” but also “invisible.” A senior engineer wrote: “I used to mentor juniors by walking them through logic. Now, the AI teaches them faster than I can. But they don’t understand why it works.” A junior wrote: “I feel like a conductor of something smarter than me. But if it disappears, I’m not sure I can perform alone.” This captures the paradox of Vibe Coding culture — empowerment through dependency. The Social Layer Perhaps most surprising was the impact on team communication. In AI-heavy teams, Slack and GitHub comments dropped by nearly 40%. Developers didn’t need to ask colleagues for help; they asked the AI instead. While this made teams more efficient, it also weakened social bonds and mentorship cycles. Several participants used the same phrase in their diaries: “It’s quieter here now.” Lessons for the Future of Vibe Coding The Dear Diary study doesn’t argue against AI. In fact, most participants wanted to keep their tools after the experiment ended. But it does raise profound questions about emotional sustainability. If AI handles the creative friction, what keeps developers engaged, challenged, and connected? The researchers suggest reintroducing human reflection layers — daily check-ins, shared code reviews, or even “AI reasoning discussions,” where developers analyze why the model made a certain choice. These small rituals restore a sense of agency and authorship, turning AI back into a partner instead of a replacement. The final line of the study reads more like a warning than a conclusion: “When creation becomes effortless, meaning becomes optional.” It’s a haunting summary of where Vibe Coding may lead — a world of limitless output, but where rediscovering purpose might become the hardest task of all.

Published on October 9, 2025