Where the reclaimed hours go

AI coding assistants have moved from novelty to near-universality in a short span. GitHub’s own survey data shows that the overwhelming majority of developers have used such tools at least once, and the State of JS 2023 survey puts regular usage at over 80%. But the more interesting question isn’t whether these tools save time—it’s what developers actually do with that time.

Across GitHub’s research and conversations with practitioners, a consistent picture emerges: the hours once lost to debugging, boilerplate setup, and documentation grunt work are being reinvested in higher-value activities. Respondents report spending more time on system design, collaboration, code reviews, and learning new technologies. The shift is less about writing more code and more about writing better code—and becoming better engineers in the process.

From trial and error to deliberate design

One of the most frequently cited benefits is the ability to spend more time in the planning and design phase. Open source developer Claudio Wunder describes spending less time on trial-and-error debugging and more time ensuring his code is secure and performant. Survey data backs this up: 40–47% of respondents say AI has enabled them to dedicate more time to designing systems and customer solutions, while 37–43% report spending more time refactoring and optimizing code.

The shift begins before a single line of code is written. Wunder uses GitHub Copilot Chat to think through project architecture—a modern twist on rubber duck debugging, except the duck talks back. Instead of merely organizing his own thoughts, he gets feedback on his ideas and starts projects with a clearer structural vision.

Generative AI for the IT Pro co-author Chrissy LeMaire echoes this sentiment. “I spend less time doing grunt work and more time just talking,” she says. For her, the appeal is that AI eliminates the tedious setup phase of a project, letting her reach the substantive work immediately.

Practical approaches to using AI for design and refactoring include:

  • State your preferences up front. Wunder begins new projects by telling Copilot Chat that he prefers ES6 built-ins and arrow functions. Explicit statements about style and conventions help steer output and make the tool’s reasoning more transparent.
  • Provide sample files. LeMaire feeds the LLM examples of similar projects, sometimes concatenating them into a single file, so the generated code and documentation match her preferred patterns.
  • Start with a skeleton function. For inline code generation, Wunder recommends defining meaningful parameters, arguments, and comments that describe what each component should do.
  • Treat debugging as a conversation. He keeps all relevant code open in VS Code and opens a new Copilot Chat session with “Let’s debug some code,” then iterates with follow-up questions about expected behavior and edge cases.

More time for teammates, not just terminals

AI’s impact extends beyond solo coding. Between 40–47% of survey respondents say AI helps them spend more time collaborating with team members, and 39–45% report more time on code reviews—a core mechanism of developer collaboration and knowledge sharing.

Part of this comes from AI-generated documentation. LeMaire describes asking an LLM to produce JSDoc-formatted documentation from JavaScript function names and parameters, achieving roughly 95% accuracy. The result was code that teammates could understand and contribute to more readily.

“I was able to go through some JavaScript and have an LLM generate JSDoc-formatted documentation based on function names and parameters with something like 95% accuracy,” she says. “My team really loved that.”

Tips for using LLMs to improve team collaboration:

  • Use existing documentation as style guides. LeMaire prompts LLMs with help text from her favorite PowerShell commands, which helps the model match the tone and format she prefers for new documentation.
  • Add header comments to every file. GitHub developer advocate Christopher Harrison recommends leaving descriptive comments at the top of each code file to help Copilot understand the codebase—and to leave useful context for teammates. He calls it “a digital version of donuts in the break room.”

Learning and experimenting on the clock

Keeping pace with new languages, frameworks, and APIs is a perennial challenge for developers—and one that AI is helping to ease. Survey results show 43–47% of respondents spending more time on learning and development, while 44–46% devote more time to research and emerging technologies.

For LeMaire, transitioning from DevOps to front-end development was made significantly less stressful by AI assistance. “Otherwise, I would have had to spend much more time context switching and looking things up,” she says.

DevOps architect Alessio Fiorentino has been using GitHub Copilot to learn Rust, a language with a steep learning curve for developers coming from Python or JavaScript. He describes AI as helping him navigate Rust’s nuances and ensuring he writes idiomatic code.

Notably, developers caution that AI tools are not a substitute for understanding. Wunder argues that even if LLMs eventually generate entire applications, developers will still need to evaluate that code. He recommends focusing on higher-level computer science concepts and sharpening communication skills—which, conveniently, are the same skills required to use AI tools effectively.

Ways to use AI as a learning aid:

  • Ask about language specifics. Wunder used Copilot to understand Go syntax, asking questions like what adding a type after :=< does, and how namespacing and module definitions work.
  • Explore unfamiliar codebases. Highlight a block of code and ask Copilot to explain it, or ask targeted questions about which variables relate to specific functionality.
  • Visualize data flow. GitHub developer advocate Kedasha Kerr has used Copilot’s mermaid diagramming features to better understand how data moves through an application.

The shifting shape of development work

AI coding tools have rapidly become a standard fixture in the development stack, and their influence on how developers spend their time is becoming clear. The trajectory points away from bug-fixing and toward design, collaboration, and continuous learning. These tools still require direction and oversight, but they are already reshaping the day-to-day reality of software development—and enabling developers to focus on the parts of their work that require genuine human judgment.