AI Code Whisperers

AI seems to be everywhere and impacting everything. Software development is no exception to this phenomena. AI coding assistants are emerging as pivotal tools designed to enhance productivity and creativity. These assistants aid developers by suggesting code, generating documentation, and facilitating debugging. As these tools integrate deeply with development environments, they offer significant time savings and improved code quality. The current state of such tools do have drawbacks as well. Understanding the strengths and weaknesses of these tools and learning how to leverage them effectively is crucial for staying ahead in a competitive tech landscape.

In preparation for a talk about AI coding assistants I am co-presenting with Arun Batchu at the Applied AI Conference - Spring 2024, I conducted an informal open-ended discussion with several experienced software engineering consultants. There were eight participants in all, including myself. The results of this discussion, summarized below with the help of ChatGPT, align well with my research on this topic.

Advantages of AI Coding Assistants

1. AI tools like GitHub Copilot are helpful for writing unit and acceptance tests, suggesting mocks, and verifying interactions with mocks.
2. ChatGPT and similar LLMs are useful for exploring concepts, ideation, and offering alternative design approaches.
3. AI coding assistants can be effective for generating documentation and explaining code.
4. They enhance test-driven development (TDD) vibes and can help debug and reason about the cause of bugs efficiently.
5. Integrated tools like Copilot can be beneficial within specific IDEs like Jetbrains products, helping to improve development workflows.
6. AI assistants can be adept at converting complex tasks (e.g., data pattern recognition or creating comprehensive documentation quickly).
7. Some developers appreciate AI's ability to provide quick feedback on coding issues and help with coding questions, especially in familiar contexts.

Challenges with AI Coding Assistants

1. AI tools frequently generate code that developers don't understand or that doesn't integrate well with existing code bases, leading to potential reliability issues.
2. The tools sometimes produce outdated or incorrect solutions, especially for newer or rapidly evolving technologies (e.g., Terraform on Azure).
3. There's a risk of AI hallucinating answers, which can lead to the use of deprecated APIs or incorrect commands.
4. Junior developers might rely too heavily on AI, potentially hindering their learning process by not engaging deeply with problem-solving.
5. Coding assistants can be inconsistent, with performance varying by tool and release, and integration with local systems can be challenging.
6. There's a concern about security and confidentiality, especially when proprietary or sensitive information might be inadvertently shared via AI platforms.
7. Overall, while AI tools can provide assistance, they often require careful management and oversight to avoid introducing errors into the code base.

Conclusion

This an exciting arena that will hopefully allow us all to spend more time on those aspects of our jobs we find most enjoyable and less in those of drudgery. We need to understand and influence their development and learn how to best use them.

According to Slashdot, there are 142 coding assistants. That is an overwhelming amount to explore. Yet, as professionals, it is an area we should be exploring. If you can, come to the conference. If you cannot attend, feel free to reach out to discuss how we can explore and learn together.