🕰️ The Oral History of Windsurf (ft. Varun Mohan, Scott Wu, Jeff Wang, Kevin Hou, Anshul R)

This Latent Space podcast episode chronicles the journey of Windsurf, from its early days as ExaFunction to its acquisition by Cognition, featuring insights from key figures involved. The podcast begins by revisiting an earlier episode with Varun Mohan, discussing ExaFunction's GPU virtualization technology and the genesis of Codium. It then transitions to discussions on AI agent capabilities, model training, and the nuances of building AI-first products, highlighting the importance of data quality and domain-specific models. The episode culminates with an exclusive interview featuring Scott Wu (Cognition) and Jeff Wang (Windsurf), addressing the circumstances surrounding the acquisition, product synergies, and future plans, while also touching upon sensitive issues and rumors in the AI community.

Outlines

Part 1: Introduction & ExaFunction Origins

Part 2: Codium's Emergence & LLM Product Building

Part 3: Devin, the AI Software Engineer

Part 4: Windsurf Editor

Part 5: Codium & Windsurf - Future Vision

Sign in to continue reading, translating and more.

Open full episode in Podwise

Latent Space

Part 1: Introduction & ExaFunction Origins

00:00Introduction to the Latent Space Podcast Special Episode: AI Engineering and Windsurf Acquisition

Introduction to the Latent Space Podcast Special Episode: AI Engineering and Windsurf Acquisition

02:17ExaFunction's Origins and Focus on GPU Efficiency

ExaFunction's Origins and Focus on GPU Efficiency

Part 2: Codium's Emergence & LLM Product Building

07:27Codium's Emergence and the Value of Code Completion

Codium's Emergence and the Value of Code Completion

12:16Data, User Growth, and the Copilot for X Concept

Data, User Growth, and the Copilot for X Concept

17:45Key Considerations for Building LLM Products

Key Considerations for Building LLM Products

22:04Code Generation Strategies and Model Size Considerations

Code Generation Strategies and Model Size Considerations

27:03Scaling Laws and the Compute Optimal Model

Scaling Laws and the Compute Optimal Model

32:34Token Limitations and the Future of Context

Token Limitations and the Future of Context

37:53LLM Ops and Model Comparison

LLM Ops and Model Comparison

42:36User Feedback, Domain-Specific Models, and Bandwidth Considerations

User Feedback, Domain-Specific Models, and Bandwidth Considerations

47:50Favorite AI Products and the Importance of Tinkering

Favorite AI Products and the Importance of Tinkering

Part 3: Devin, the AI Software Engineer

52:41Introduction to Devin, the AI Software Engineer

Introduction to Devin, the AI Software Engineer

54:08Codium's AI Developer Tools and Focus on Context Awareness

Codium's AI Developer Tools and Focus on Context Awareness

55:32Context Retrieval Methods and the Limitations of Embeddings

Context Retrieval Methods and the Limitations of Embeddings

1:00:04Codium's Approach to Retrieval and the mQuery System

Codium's Approach to Retrieval and the mQuery System

1:04:04Codium's Vertical Integration and the Results of mQuery

Codium's Vertical Integration and the Results of mQuery

1:07:28Future Applications and the Analogy to Autonomous Driving

Future Applications and the Analogy to Autonomous Driving

1:11:03Call to Action and Introduction to Devin

Call to Action and Introduction to Devin

1:12:36Demo of Devin Building a Mobile-Friendly Website

Demo of Devin Building a Mobile-Friendly Website

1:17:14The Vision Behind Devin and the Agentic Workflow

The Vision Behind Devin and the Agentic Workflow

1:20:05The UX Problem and Features of Devin

The UX Problem and Features of Devin

1:22:35How Devin is Changing Workflows

How Devin is Changing Workflows

1:25:01The Future of Software Engineering with AI

The Future of Software Engineering with AI

1:27:31Q\&A: Access to Devin

Q\&A: Access to Devin

1:29:00Q\&A: Impact on Junior Engineers and Challenges to Realizing the Vision

Q\&A: Impact on Junior Engineers and Challenges to Realizing the Vision

Part 4: Windsurf Editor

1:31:03Launching the Windsurf Editor

Launching the Windsurf Editor

1:31:41AI Flow and the Core Motivation of the Windsurf Editor

AI Flow and the Core Motivation of the Windsurf Editor

1:32:31Knowledge and Tools in the Windsurf Editor

Knowledge and Tools in the Windsurf Editor

1:33:40Human Action and the Goal of Ambitious Projects

Human Action and the Goal of Ambitious Projects

1:34:47Future Potential and Features of the Windsurf Editor

Future Potential and Features of the Windsurf Editor

1:36:11The Magic of the Windsurf Editor

The Magic of the Windsurf Editor

Part 5: Codium & Windsurf - Future Vision

1:37:20Introduction to the Latent Space Podcast with Codium

Introduction to the Latent Space Podcast with Codium

1:38:32Progress Since Last Appearance and the Launch of Windsurf

Progress Since Last Appearance and the Launch of Windsurf

1:42:04Limitations of VS Code and the Vision for Windsurf

Limitations of VS Code and the Vision for Windsurf

1:44:45The Importance of Evals and the Human Element

The Importance of Evals and the Human Element

1:47:00The Importance of Incomplete Code

The Importance of Incomplete Code

1:49:27The Importance of Data