How AI Will Change the Way Developers Work (Tabnine’s Vision Explained)

Capabilities

Peter Guagenti, President and Chief Marketing Officer at Tabnine, explains the extensive capabilities of AI code assistants. These tools go beyond code generation to include break fix, documentation, and autonomous testing, offering a spectrum from AI assistants to fully autonomous AI software engineers 1. Guagenti emphasizes the significant productivity gains, with Tabnine's ROI calculator showing potential savings of $50,000 to $70,000 per engineer annually 2.

The ROI is sort of clear. The question is not ROI, but where and how do you deploy it?

--- Peter Guagenti

These savings can be redirected to reduce tech debt and enhance application development.

   

Autonomy

Guagenti discusses the current state and future of autonomous code generation. He notes that while some tools can perform tasks autonomously, the goal is to extend these capabilities across the entire software development lifecycle (SDLC) 3. The focus is on automating maintenance, break fix, refactoring, and security reviews to improve efficiency 4.

We're working through the entire SDLC... moving into code review, security review, and other areas.

--- Peter Guagenti

This approach aims to ensure that AI-generated code meets company standards and passes pull requests.

   

Code Quality

Ensuring code quality is a critical aspect of using AI in software development. Guagenti highlights the importance of code reviews to maintain high standards and prevent the introduction of low-quality code into training datasets 5. He stresses the need for strict curation of training data to ensure AI models are trained on high-quality sources 6.

We're going to have to get more and more strict about what it is we're actually training on.

--- Peter Guagenti

This meticulous approach helps in maintaining the integrity and performance of AI-generated code.

   

Productivity

Tabnine's AI tools have significantly improved productivity in software development teams. Guagenti shares that Tabnine has contributed to writing 1-2% of the world's code, with AI generating 30-50% of code in projects it touches 7. He cites third-party studies showing 20-25% productivity savings for software engineering teams 8.

We've probably written one to 2% of all of the world's code at this point.

--- Peter Guagenti

These improvements allow teams to focus more on creative and complex tasks, enhancing overall efficiency.