AI Revolutionizing Development Productivity

Motivation

The adoption of AI in software development is effectively evolving our work methods, unlocking numerous possibilities to boost productivity and streamline operations. This week, we'll examine some crucial areas where AI is making a substantial influence. This is particularly relevant as we're investing in similar initiatives in OneDrive and SharePoint. In our exploration, we'll be referencing and building upon research conducted by global AI community in this field.

The Impact of AI on Coding Speed?

• Recent studies, including one involving Google software engineers, have demonstrated that AI-powered tools can substantially reduce the time developers spend on coding tasks. The study found that developers using AI were approximately 21% faster on average, compared to those who did not.? ?
• This speed increase is not uniform, however. Developers who spend more hours coding daily tend to see a greater benefit from AI assistance. ?
• Other research indicates even more dramatic results, with one study reporting a 56% increase in coding speed using AI code assistants such as GitHub Copilot. ?
• A separate study observed a 26% increase in developer throughput (measured by pull requests) when using AI tools.

These findings highlight the significant potential of AI to accelerate our development cycles and allow us to deliver features more quickly. However, it's worth noting that these figures can vary due to the specific AI tools used and the context.

Let's explore three innovative ways AI can supercharge productivity in the field of software development.

A. AI-Powered tools / features for writing new code

Several AI-enhanced features are making a real difference in how code is written:

• AI Code Completion: This technology provides real-time code suggestions as developers type, significantly speeding up the coding process. It predicts and completes lines of code based on context. This is similar to the autocomplete that many of us are used to, but it is more advanced due to leveraging a large language model. ?
• Smart Paste: Another useful feature uses AI to automatically adjust code snippets pasted from one area to another in the IDE, making integration more seamless. ?
• Natural Language to Code: This allows developers to describe what they want in natural language, and the AI will generate the corresponding code, bridging the gap between idea and implementation. This ties into our text-to-code conversion project, with the potential to enhance the speed at which we transform our visions into reality.

These features, particularly the code completion and natural language to code tools, directly align with our projects, offering the chance to accelerate both our development and debugging processes.

B. AI for Code Migration and Refactoring

AI isn’t just about writing new code, it's also proving its worth in modernising existing systems.

• LLMs are being used to handle complex code migrations, such as converting code from older to newer versions, with impressive efficiency. The Google engineers successfully migrated code with a 50% reduction in migration time using AI-based tooling. ?
• AI can handle tedious tasks like refactoring code, updating libraries, and cleaning up experimental features, which can take up considerable time when done manually. ?
• These AI-powered migrations often require a mix of LLMs with AST-based techniques and heuristics. The goal is to find opportunities where LLMs can reduce the need for difficult-to-maintain Abstract Syntax Tree (AST)-based transformations. A common toolkit, incorporating LLM prompts, AST, and symbol-based techniques is used across migrations to streamline the process, demonstrating adaptability.

The above shows that AI’s capabilities go beyond code writing and offer an opportunity to tackle substantial technical debt that slows down teams and affects development speed.

C. Agent-Based Program Repair

Beyond code generation and migration, AI is advancing into automated bug fixing.

• Agent-based program repair systems can autonomously analyse bugs, generate tests, localise the fault, and propose fixes. ?
• One system, called Passerine, was able to generate plausible patches for 73% of machine-reported bugs and 25.6% of human-reported bugs in a dataset of Google issues. ?
• These systems leverage LLMs to understand the code context and apply fixes, working within an enterprise’s unique development environment. ?
• Importantly, AI agents perform better when provided with more detailed bug reports, implying better reporting practices are increasingly relevant. ?
• They can even adapt their behaviour based on the type and information provided in bug reports.

These results suggest a future where AI takes a more active role in bug resolution, allowing us to allocate our efforts to more intricate challenges.

Challenges and Considerations

Despite the immense progress, it's crucial to remain mindful of the challenges:

• Context limitations: Large codebases and lengthy logs can sometimes exceed the context window of AI models, which requires innovative solutions. ?
• Tool limitations: AI tools are still evolving, they may fail or attempt to use non-existent commands.

Therefore, while we should embrace AI, we must also understand its limitations and be prepared to adjust our workflows as we integrate these tools into our everyday work.

Conclusion

AI is not just a trend; it's a fundamental shift in how software is developed. By adopting AI-powered tools for code completion, migration, and bug fixing, we can significantly enhance our productivity, reduce our workloads, and accelerate our time-to-market. As we continue to explore and integrate these technologies, we will be well-positioned to leverage AI to its full potential. Our team’s projects in hard coding detection, text-to-code, and code completion are very well aligned with these trends, and we look forward to continuing to explore the opportunities that these bring. Do you have any questions or thoughts about these AI trends? Any specific use cases in mind? Let's continue this AI conversation and shape the future of software development together.

?

Ref:??

1. How much does AI impact development speed? An enterprise-based randomized controlled trial
2. Evaluating Agent-based Program Repair at Google
3. How is Google using AI for internal code migrations