The Promise of Hyper-Productivity
Tools like GitHub Copilot entered the scene with the promise of a revolution. By generating code snippets, translating languages, and handling repetitive boilerplate, these AI assistants were positioned to supercharge efficiency. The pitch was simple:
by offloading mundane tasks, developers could conserve their mental energy for more complex, creative, and satisfying problem-solving. Early studies seemed to confirm this, with some showing developers completing tasks significantly faster. Companies embraced the technology, with Gartner predicting that by 2028, 75% of enterprise software engineers will use AI code assistants, a massive leap from less than 10% in early 2023. The goal was clear: to build better software, faster.
A New Productivity Paradox
Despite the hype, the reality on the ground is proving more complex. Several studies and developer anecdotes point to a 'productivity paradox': while individual coding tasks may get faster, the overall time to deliver a finished feature is not shrinking—and in some cases, it’s growing. One 2025 study found that experienced developers using AI tools actually took 19% longer to complete tasks, even though they believed they were working faster. The issue isn't that the tools don't work. It's that they change the nature of the work itself, creating new bottlenecks and new types of labor that are often overlooked. The time saved on typing is often spent on other activities, such as prompting the AI, waiting for responses, and, most critically, reviewing the code it generates.
When More Becomes the New Normal
As developers began producing code at a faster clip, a predictable human element kicked in: expectations rose. Seeing the increased velocity of individual tasks, some managers and organizations have simply expanded the scope of what's considered a reasonable workload. This creates a treadmill effect. The 10x output of one developer can become a 10x review burden for another. This is a classic local optimization problem, where speeding up one part of a system overwhelms another. The result is that while developers are busier and generating more code, the overall flow of value to the customer doesn't necessarily accelerate. Instead, developers find themselves juggling more workstreams and facing a ballooning queue of code to review, leading to a different kind of burnout rooted in volume, not just difficulty.
The Hidden Cognitive Load
Working alongside an AI assistant introduces a new form of cognitive strain. Developers must now constantly validate the AI's output, which can be subtly wrong. A 2026 report noted that 96% of developers have trouble trusting AI-generated code. This shift turns the developer from a creator into a supervisor, a role that demands intensive oversight and can be mentally exhausting. Some have described this as trading typing time for reading and untangling time. Furthermore, the constant context-switching between writing code, crafting prompts, and reviewing suggestions can break a developer's 'flow state'—the deep concentration required for complex problem-solving. The pressure to keep up with the dizzying pace of new AI tools also adds to the stress, creating a sense of 'AI FOMO' (fear of missing out).
Finding a Sustainable Path Forward
The solution isn't to abandon these powerful new tools, but to integrate them more thoughtfully. Experts suggest companies should focus on holistic team productivity rather than just individual output metrics like lines of code. This involves making bottlenecks visible across the entire development lifecycle, from requirements gathering to final deployment. It also requires providing clear training and establishing feedback loops so that developers feel supported rather than overwhelmed. For developers themselves, it may mean setting boundaries and intentionally protecting 'craft hours' for coding without AI assistance to maintain their skills and a sense of connection to their work. As one developer noted, the goal should be to use technology to strengthen people, not just processes.
