Real-world study reveals how developers actually use AI coding assistants—and when mixing tools backfires

A new empirical study of professional developers working in their natural environments reveals nuanced insights into how they interact with generative AI coding tools—moving beyond simple productivity metrics to examine developers' well-being, cognitive load, and optimal tool usage patterns.

Study design and key findings#

Researchers conducted a mixed-methods investigation that combined controlled data collection with observation of developers' everyday work. The study found that developers generally embrace AI assistants, particularly for monotonous, repetitive, and structured coding tasks, and perceive both efficiency and productivity benefits overall.

However, the research uncovered a counterintuitive pattern: developers benefit most when they stick to a single interaction mode per task. When combining in-code suggestions with chat-based prompting within the same coding task, the advantages diminish—suggesting that mode-switching introduces friction rather than flexibility.

Task-specific preferences and interaction modes#

The research identified distinct preferences based on task characteristics. Developers leverage in-code suggestions and chat-based prompting differently depending on the work at hand. Each mode independently improves task efficiency and reduces perceived workload, but their combination within a single task undermines these gains.

Based on these patterns, the researchers propose a practical rule of thumb for choosing the right interaction type based on what the developer is trying to accomplish.

Cognitive load vs. output quality#

During development-heavy work, an important distinction emerged: perceived cognitive load comes primarily from managing the AI interaction itself, while developers' sense of productivity depends on the quality of the AI's output. This separation suggests that seamless tool integration matters as much as generating correct code.

Study impact and broader implications#

Participation in the research itself shifted developers' behavior—those involved became more aware and intentional in their use of GenAI tools going forward. This finding underscores the value of real-world field studies that capture authentic developer experiences rather than relying on laboratory settings or retrospective surveys alone.