The AI Productivity Promise
AI coding assistants like GitHub Copilot and Amazon Q Developer are now used by a vast majority of software engineers. The pitch was irresistible: these tools, powered by large language models, can generate boilerplate code, suggest entire functions,
and even translate code between languages in an instant. Initial studies backed this up, with one prominent experiment showing developers completed a task 55.8% faster with an AI assistant. The expectation was clear—by automating the tedious and repetitive parts of coding, developers would save time, reduce frustration, and perhaps even log off earlier. This has led to widespread adoption, with some surveys indicating that over 80% of developers now use or plan to use these tools.
Faster Code, Higher Expectations
The problem with moving faster is that the goalposts often move, too. When developers can build features more quickly, the business tends to expect more features in the same amount of time. Instead of a shorter workday, the pace of work simply accelerates. This phenomenon is a classic case of what’s known as Jevons paradox—as efficiency increases, so does demand. The time saved on individual coding tasks doesn't necessarily translate into more free time. Instead, it gets reinvested into a higher volume of work. As a result, the pressure to deliver doesn't decrease; it just shifts to a new, faster baseline, which can contribute to burnout rather than prevent it. A 2025 DORA report noted that increased AI use correlated with a drop in delivery throughput, suggesting that more AI-generated code doesn't automatically lead to healthier systems.
The Hidden Cost of AI-Generated Code
Writing code is only one part of a developer's job. In fact, some analyses show that developers spend as little as 16% of their time writing new code. The rest is spent on debugging, testing, collaborating, and reviewing code written by others. AI assistants are creating a new bottleneck in this process: verification. While these tools write code quickly, it's not always perfect or secure. One study found bug rates increased by 41% for developers with access to Copilot. Developers report spending significant time reviewing, cleaning up, and modifying AI-generated suggestions. Trust in AI output has actually fallen, even as adoption has risen. The cognitive load of carefully vetting code you didn't write yourself is substantial, and in some cases, it can even slow experienced developers down. A 2025 study found that seasoned developers took 19% longer to complete tasks with AI tools, largely due to the overhead of reviewing and correcting the output.
A Shift in the Developer's Role
AI isn't just a tool; it's changing the very nature of the job. As AI handles more of the raw code generation, the developer's role is evolving from a writer to an editor, curator, or system architect. This shifts the focus from low-level implementation details to high-level problem-solving and system design. One Harvard study found that developers using Copilot spent 12.4% more time on core coding activities but also saw a significant decrease in time spent on project management and peer collaboration. While this allows developers to focus on more stimulating work, it also means they are spending more of their day on the most cognitively demanding tasks. The easy, almost mindless parts of the job that once provided a mental break are being automated away, leaving a workday filled with complex challenges that can be more draining and, ultimately, longer.
