At the recent DeveloperWeek conference in Santa Clara, California, software developers shared their experiences with AI coding assistants, acknowledging their potential to accelerate software development, but also highlighting their limitations and areas for improvement.

According to attendees, AI coding tools, such as GitHub Copilot, Tabnine, and JetBrains AI Assistant, are already proving to be useful, but still need to mature. Jens Wessling, CTO and chief architect at software security company Veracode, noted that these tools "certainly provide an opportunity to accelerate software development," but added that it will be interesting to see how they address issues like security and correctness in the long term.

Many developers at the conference expressed optimism about the potential of AI coding assistants. Juan Salas, CTO at Alto, which provides software development services, said that these tools are "great as tools" and can help save time if users know how to use them. College student Aasritha M. praised GitHub Copilot's ability to recognize patterns in code and anticipate what the developer is about to do next, although she noted that ChatGPT does a better job of finding mistakes in code.

Other developers, such as Sahil Shah and Ratna Maharjan, also found GitHub Copilot and ChatGPT to be "pretty helpful" in providing code snippets and accelerating their work. However, they also expressed some dissatisfaction with the tools. Wessling noted that AI coding tools often struggle with libraries and library versions, and occasionally provide "random answers." Shah added that the answers GitHub Copilot gives him are sometimes vague.

Developer and retired physicist Peter Luh tested GitHub Copilot on four math problems during the conference and reported that it failed miserably on all four. However, he believes that Copilot might be suitable for general chats that include "hallucination" responses.

Despite the limitations, many developers believe that AI coding assistants will improve over time. Salas emphasized the importance of combining AI with human direction, noting that AI coding assistants can give an illusion of getting to a solution quickly, but often require more technical nuance and knowledge of what to ask them. He believes that AI coding assistants will get better, but users need to be aware of their limitations and use them effectively.

The feedback from developers at the conference highlights the ongoing evolution of AI coding assistants. While they show promise in accelerating software development, they still need to mature and address issues like security, correctness, and consistency. As the technology continues to improve, it will be interesting to see how developers adapt and integrate AI coding assistants into their workflows.

In the meantime, developers are encouraged to experiment with AI coding tools and provide feedback to help shape their development. With continued innovation and refinement, AI coding assistants have the potential to revolutionize the software development process and unlock new efficiencies and productivity gains.