Exploring Sourcegraph's Levels of Code AI: A New Framework for Software Development
Welcome to my talk on the levels of Code AI, where we're building Cody, the only AI coding assistant that knows your entire codebase.
- 1. Otto Cookit, Director of Developer Relations at Sourcegraph, talks about their concept of "levels of code AI."
- 2. Code AI refers to software that builds software; currently, 92% of developers use code AI tools, up from just 1% a year ago.
- 3. Sourcegraph's founder and CEO, Quinn Slack, predicted that in 5 years, 99% of code will be written by AI.
- 4. There are six levels of code AI, divided into three categories: human-initiated (where humans are the primary coders), AI-initiated (where AI starts to take a proactive role), and AI-led (where AI
- 5. Level 0: Developers write all code manually without any AI assistance; IDE features like symbol name completion provide minimal help. This reflects traditional software development before AI was in
- 6. Level 1: Developers use AI that generates single lines or blocks of code based on developer intent; AI has been trained on millions of open-source code lines to provide superior completions based o
- 7. Level 2: The AI coding assistant understands the specific codebase, allowing for better suggestions for code completions tailored to the current project.
- 8. Levels 0-2 are similar to SAE levels of autonomy for vehicles, with full human control and assistance from AI.
- 9. Level 3: Developers provide high-level requirements, and the AI assistant delivers a code-based solution, building out full components and integrations.
- 10. At SAE level 3, vehicles take on the primary role of driving, with humans as fallbacks; similarly, level 3 code AI can handle coding tasks without developer oversight.
- 11. Level 4: The code AI assistant proactively handles coding tasks autonomously; it could monitor code changes and submit pull requests for documentation or bug fixes.
- 12. At SAE level 4, vehicles can perform virtually all driving tasks under specific conditions, with humans as a fallback; similarly, level 4 code AI assists developers in maintaining their projects w
- 13. Level 5: The AI assistant requires minimal guidance and handles the entire software development life cycle from architecture design to deployment and continuous improvement.
- 14. At SAE level 5, vehicles perform all driving tasks under all conditions with no human intervention; level 5 code AI can take code from concept to production without human interaction.
- 15. Sourcegraph believes these six levels provide a useful framework for understanding the evolution of code AI and its impact on software development.
- 16. They welcome feedback and discussion on this topic, particularly at the AI Engineer Summit.
- 17. The presentation aims to educate customers and users while also shaping Sourcegraph's thinking about code AI.
- 18. Code AI is a rapidly evolving field, with significant potential to change how software development is done.
- 19. Understanding these levels can help developers anticipate and adapt to the changes brought by code AI.
- 20. Code AI has the potential to increase efficiency, reduce errors, and improve overall software quality.
- 21. As code AI evolves, it's crucial for developers to remain informed about its capabilities and limitations.
- 22. Discussions like this one help advance the field of code AI and ensure its responsible adoption in software development.
- 23. By engaging with the developer community, Sourcegraph hopes to foster a better understanding of code AI and its potential impact on the industry.
- 24. The presentation concludes by inviting the audience to visit booth G5 for further discussion and encouraging them to try Cody, Sourcegraph's AI coding assistant, in their preferred IDE.
Source: AI Engineer via YouTube
❓ What do you think? What are your thoughts on the ideas shared in this video? Feel free to share your thoughts in the comments!