---
source: hn
url: https://www.scientificamerican.com/article/why-developers-using-ai-are-working-longer-hours/
published_at: '2026-03-07T23:39:04'
authors:
- birdculture
topics:
- ai-for-coding
- developer-productivity
- software-engineering
- burnout
- code-quality
relevance_score: 0.89
run_id: materialize-outputs
language_code: en
---

# Why developers using AI are working longer hours

## Summary
This article synthesizes multiple industry reports and studies, pointing out that although AI coding tools often increase individual output, they are also often accompanied by longer working hours, higher release instability, and potential skill degradation. The core conclusion is: AI has not automatically reduced the human labor involved in software engineering; instead, it may shift pressure, rework, and cognitive costs onto developers.

## Problem
- The article examines the question: **why developers often do not become more relaxed after using AI, but instead work longer and feel more pressure**.
- This matters because software engineering is one of AI’s most promising application scenarios; if “higher efficiency” also brings **more rollbacks, more overtime, skill degradation, and burnout risk**, then the real benefits to companies may be overestimated.
- For the field of code intelligence and automated software production that users care about, this means that **generating code is not the same as completing software engineering**; verification, debugging, collaboration, and organizational management remain key bottlenecks.

## Approach
- This is not a single experimental paper, but rather a **synthetic analysis based on multiple surveys, corporate studies, and academic reports**, summarizing AI’s actual impact on software development.
- The core mechanism is simple: **AI makes writing code faster, so organizations and individuals try to do more within the same amount of time; but software still requires humans to verify, debug, and adapt it to special needs, so the time saved is often filled by new tasks, rework, and higher expectations.**
- The article breaks the impacts into several categories: **higher individual productivity**, **increased software delivery instability**, **spillover of work time**, **damage to skill learning and debugging ability**, and **changes in collaboration structures**.
- The article also emphasizes an organizational-level amplifier: in the context of layoffs and efficiency-driven management, AI is often used to “make fewer people do more work,” so the pressure may not come from the model itself, but from **management expectations and evaluation mechanisms**.

## Results
- A Google DORA survey of **nearly 5,000 technology professionals** found that **90%** of respondents said they used AI at work, and **more than 80%** said AI improved productivity; however, at the same time, as AI use increased, **software delivery instability** also rose, and developers were **more likely to roll back changes already deployed online**.
- An evaluation by Multitudes of **more than 500 developers** found that engineers merged **27.2% more pull requests** on average, but **out-of-hour commits increased by 19.6%**, indicating that expanded output was accompanied by spillover into working hours and potential burnout risk.
- A February 2025 report from UC Berkeley Haas / Harvard Business Review said that at a U.S. technology company, after employees adopted AI, they **took on more tasks, worked at a faster pace, and worked longer hours**; the article did not provide specific percentages, but clearly noted that employees began using AI during **lunch breaks, rest periods, and meetings** as well.
- A January 2025 report from Anthropic found that when completing tasks using a new software library, the AI-assisted group showed only a **small and statistically insignificant speed improvement** relative to the control group; but on a post-task test, the AI-assisted group scored **17% lower**, with the largest gap on **debugging-related questions**.
- The article also cites a 2025 Harvard Business School working paper: AI may cause open-source developers to shift time away from **project management / code review / issue maintenance** toward **directly generating code themselves**; the article does not provide specific numbers, but this change is considered likely to weaken opportunities for junior developers to learn through collaboration and build networks.
- The strongest overall conclusion is not that “AI is ineffective,” but rather: **AI improves local coding efficiency, yet does not eliminate verification, debugging, collaboration, or organizational pressure, and therefore may simultaneously increase output, rework rates, and overtime risk overall.**

## Link
- [https://www.scientificamerican.com/article/why-developers-using-ai-are-working-longer-hours/](https://www.scientificamerican.com/article/why-developers-using-ai-are-working-longer-hours/)