The article further discusses the application of Language Model (LLM) to software development, with LLM-based inline code completion being the most popular application. The authors note that AI-based suggestions have transformed the role of the code author into a reviewer, emphasizing the importance of balancing the cost of review and added value. Looking ahead, the article predicts that the next wave of benefits will come from ML assistance in a broader range of software engineering activities, such as testing, code understanding, and code maintenance. The authors also highlight the need for common benchmarks to help move the field towards practical engineering tasks.
Key takeaways:
- Google has been using AI and machine learning to improve software development, with a significant number of engineers using ML-based autocomplete tools for code completion.
- They have found that the most successful AI-based features are those that naturally blend into users' workflows, and that quick iterations with online A/B experiments are key to improving these features.
- High-quality data from the activities of Google engineers across software tools is essential for improving the quality of their models.
- Looking forward, Google plans to double down on using the latest foundation models to power existing and new applications of ML to software engineering, with a focus on tasks such as testing, code understanding, and code maintenance.