[1/2] AI in software engineering at Google: Progress and the path ahead (AI column)
It's been great this summer. posture from the guys at Google about their approach to using AI in engineering. The highlights of this article are as follows:
- The author works in a team responsible for the software development environments in which Google engineers spend most of their time, including the inner loop. (IDE, code review , code search)and the surface of the outer loop (For example, error management, planning).
- The use of AI is now ubiquitous and affects developer productivity and satisfaction, and is measured using AI. den
- The guys have three guiding principles they use when embedding AI in engineering products.
- Prioritize by technical feasibility and impact: Work on ideas wherein both technical feasibility has already been established and high (measureable) impact on engineers’ workflows is expected.
- Learn quickly, to improve UX and model quality: Focus on iterating quickly and extracting lessons learned, while safeguarding developer productivity and happiness. User experience is just as important as model quality.
- Measure effectiveness: As our goal is to increase productivity and satisfaction metrics, we need to extensively monitor these metrics.
Productivity development was well described in the articles
- "Developer Productivity for Humans, Part 7: Software Quality-I handler earlier
- "Measuring Developer Goals-I handler earlier
- The scheme of applying LLM to software looks like this: Engineers use AI tools Further, the interaction logs are used for tuning features for the real needs of engineers. Historical logs are used to train AI models and then we go back to the beginning.
- Interestingly, the AI assistant's suggestions for writing code have risen to 37%, and by the amount of code created to 50% (number of symbols). You can read more about what LLM is used for in engineering activities in the previous post.Large sequence models for software development activities"
Continuation of the review in next post.
#Software #AI #ML #Engineering #Processes #DevEx