Michael Truell: Building Cursor at 23, Taking on GitHub Copilot, and Advice to Engineering Students

Key Concepts

• AI-powered coding
• Code automation
• Software development evolution
• AI copilots
• Model training
• Data scraping
• End-to-end encryption
• Product pivoting
• Codebase awareness
• Word-of-mouth growth
• AI-assisted coding workflow
• Future of coding skills

Origin Story and Early Projects

Michael Trello's journey began in middle school with an interest in starting a company. His initial foray into programming was sparked by the desire to create a mobile game with his brother. Encountering Objective-C in Xcode proved challenging, leading his brother to abandon programming, but Michael persisted, learning Objective-C and developing mobile games. He was also inspired by essays from Paul Graham (PG) and Sam Altman from Y Combinator (YC).

• Mobile Game Attempt: Initial programming experience involved trying to create a mobile game using Xcode and Objective-C.
• Spoofing App: Created a mobile app to spoof high scores in games like Piano Tiles and Flappy Bird, which went viral. This highlighted the importance of simplicity and market appeal over technical complexity.
• Robotic Dog Project: With a friend, he aimed to build a robot dog that could be trained through positive and negative feedback, similar to training a real dog.
• Reinforcement Learning (RL) Exploration: The robotic dog project led to exploring genetic algorithms, neural networks, and reinforcement learning. They focused on making RL algorithms more data-efficient, learning from limited and noisy data.
• Robot Arm and Kiwi Drive Robot: Built a multi-axis robot arm that could play ping pong and a Kiwi drive robot that could follow a line, using their own implemented neural networks due to microcontroller limitations.
• Custom Neural Network Implementation: Implemented a tiny neural network library from scratch due to the constraints of microcontrollers, providing a deep understanding of the fundamentals.

Founding Anysphere and the Pivot to Cursor

After graduating from MIT in 2022, Michael and his co-founders started Anysphere. Their initial idea was a co-pilot for mechanical engineers, using AI to predict actions in CAD systems like SolidWorks and Fusion 360.

• Mechanical Engineering Co-pilot: First major project involved building an AI co-pilot for mechanical engineers, predicting actions in CAD software.
• Data Scraping Challenges: Faced challenges in scraping CAD models from various fragmented systems and converting them into a canonical format.
• End-to-End Encrypted Messaging System: Two co-founders worked on an end-to-end encrypted messaging system that hid metadata about who was communicating with whom, addressing privacy concerns in apps like Signal and WhatsApp.
• Lack of Traction: Both the CAD co-pilot and the messaging system failed to gain significant traction, leading to a re-evaluation of their direction.
• Inspiration from GitHub Copilot: Inspired by GitHub Copilot, they initially avoided AI coding tools due to perceived competition.
• Realization of Coding Automation Potential: Recognizing the potential for AI to fundamentally change coding, they decided to focus on building a tool that automates coding and transforms software development.

Building Cursor: From Scratch to VS Code Foundation

The team pivoted to building Cursor, an AI-powered code editor. They initially built their own editor from scratch before switching to a VS Code foundation.

• Initial Editor from Scratch: Built their own code editor from scratch, utilizing open-source components like CodeMirror and language servers.
• Remote SSH and Copilot Integration: Developed their own remote SSH and Copilot integration.
• First Public Release: Shipped the first public version of Cursor approximately three months after starting development.
• Early AI Feature Iteration: The first AI feature was a universal remote that allowed users to ask the AI to perform various tasks.
• Switch to VS Code Foundation: Realized the complexity of building a complete code editor and switched to basing Cursor on VS Code, similar to how browsers use Chromium's rendering engine.
• Inspiration from Codecs: Inspired by the Codex paper, which detailed the autocomplete model behind GitHub Copilot.
• Model Training: Initially avoided training their own models but later found it to be a crucial product lever, especially as they scaled.

Growth and Future Vision

Cursor experienced significant growth, driven by improvements in the product and word-of-mouth adoption. The company focused on making the product better, leading to compounding growth.

• Growth in 2023: Experienced growth in 2023, but the numbers were still relatively small.
• Focus on Product Improvement: Focused on making the product better, leading to immediate improvements in growth metrics.
• Codebase Awareness and Next Action Prediction: Key improvements included making Cursor codebase aware and enabling it to predict the user's next action.
• Word-of-Mouth Growth: Growth was primarily driven by word-of-mouth, with users spreading the word about the product.
• YC Batch Adoption: Saw a significant increase in adoption among YC companies, with Cursor becoming the preferred tool for many builders.
• Twitter Evangelism: Early evangelism on Twitter by a co-founder helped kickstart initial adoption.
• Company Size in 2023: Remained a small team, with fewer than 10 people by the end of 2023.
• Future of Coding: Believes that AI will transform software development over the next couple of decades, with a long messy middle where engineers work alongside AI.
• Importance of Programming Skills: Emphasizes that programming remains a valuable skill, providing a good general education and practical knowledge.
• Advice for Aspiring Founders: Recommends working on projects that you are interested in with people you respect and enjoy being around.

Notable Quotes

• "We realized we were really inherently excited about the future of coding...there was going to be an opportunity for all of coding to change in the next 5 years and for all of software development to flow through models."
• "It felt like no one working on the space at the time was really taking that seriously."
• "If you make the best thing, people hear about it and talk about it."

Technical Terms and Concepts

• Objective-C: An object-oriented programming language used primarily for Apple products.
• Xcode: An integrated development environment (IDE) for macOS, used for developing software for Apple platforms.
• Genetic Algorithms: Optimization algorithms inspired by the process of natural selection.
• Neural Networks: Machine learning models inspired by the structure and function of the human brain.
• Reinforcement Learning (RL): A type of machine learning where an agent learns to make decisions by interacting with an environment and receiving rewards or penalties.
• Microcontrollers: Small, low-power computers used to control electronic devices.
• TensorFlow and PyTorch: Open-source machine learning frameworks.
• CAD (Computer-Aided Design): Software used for creating, modifying, analyzing, and optimizing designs.
• SolidWorks and Fusion 360: Popular CAD software applications.
• LMS (Learning Management System): A software application for the administration, documentation, tracking, reporting, automation, and delivery of educational courses, training programs, or learning and development programs.
• Codecs: A paper detailing the autocomplete model behind GitHub Copilot.
• Language Servers: Provide code completion, error checking, and other language-specific features to code editors.
• AGI (Artificial General Intelligence): A hypothetical level of AI that can perform any intellectual task that a human being can.

Synthesis/Conclusion

Michael Trello's journey with Cursor highlights the importance of identifying a compelling vision, iterating through multiple ideas, and focusing on product excellence. The initial focus on mechanical engineering, while ultimately unsuccessful, provided valuable experience in data scraping and model training. The pivot to AI-powered coding, driven by a belief in the transformative potential of AI, led to significant growth and adoption. The decision to build Cursor on top of VS Code allowed the team to focus on AI-specific features and improvements. The success of Cursor demonstrates the power of word-of-mouth growth and the importance of continuously improving the product based on user feedback. Trello's advice to work on projects you are passionate about with people you respect underscores the importance of intrinsic motivation and collaboration in building a successful company.