![cover of episode Scott Wu - Building Cognition - [Invest Like the Best, EP.401]](https://megaphone.imgix.net/podcasts/b296237a-bc20-11ef-b2ab-f38e0030a7f6/image/72014bf22547e920bf2efc062163845d.jpg?ixlib=rails-4.3.1&max-w=3000&max-h=3000&fit=crop&auto=format,compress)
Scott Wu - Building Cognition - [Invest Like the Best, EP.401]
Invest Like the Best with Patrick O'Shaughnessy
Deep Dive
What is Cognition's business and what does Devin do?
Cognition is an applied AI lab that has created Devin, the first fully autonomous software engineer. Devin can handle complete engineering workflows, from bug fixing to submitting pull requests, and functions at the level of a junior software engineer. It integrates into tools like Slack, GitHub, and works in the local developer environment.
How does the level of sophistication of Devin compare to human engineers?
Devin is currently at the level of a junior engineer. Initially, it was like a high school CS student, then an intern, and now it's comparable to an entry-level junior engineer. However, it can't handle complex tasks like re-architecting an entire codebase.
What is the pricing model for Devin and why is it different from other AI tools?
Devin is priced at $500 a month for engineering teams. This is based on usage, with a unit called ACUs (agent compute units) that measure the decisions and work Devin does. The pricing reflects the value of Devin's capabilities, which can make tasks 10x cheaper to complete, compared to a more general $20 per month model for text-based tasks.
What are the key differences between IDE assistants and Devin's approach?
IDE assistants provide text completion and simple code suggestions, improving productivity by 10-20%. Devin, however, functions as a full co-worker, handling tasks asynchronously and end-to-end, from debugging to testing and submitting pull requests. This can save 90% of the time on routine tasks.
How does Scott Wu describe the history of computer programming?
Scott describes programming's history as a growing abstraction layer that simplifies tasks and reduces the need for deep technical knowledge. From punch cards to assembly, BASIC, C, and modern languages like Python, each step has made programming more accessible, but the core remains about telling the computer what you want it to do.
What qualities make a 10x engineer in software development?
A 10x engineer deeply understands abstractions and can effectively solve problems by thinking through customer needs and designing efficient solutions. While they only spend about 10% of their time on creative problem-solving, this quality distinguishes them from average engineers.
What are the most surprising use cases for Devin that Scott has observed?
One of the most surprising use cases for Devin is its ability to handle large-scale migrations, modernizations, and version upgrades. These tasks are often tedious and time-consuming, but Devin can complete them much faster, saving engineers significant time and effort.
Why does Scott believe AI will surpass the best competitive programmer in 1-2 years?
Scott believes AI will surpass the best competitive programmer, like Gennady Korotkevich from Belarus, within 1-2 years. Current AI models are already competitive with top human programmers and are rapidly improving. This milestone will be a significant moment, similar to AlphaGo, demonstrating the potential of AI in programming.
How does Scott envision the future of software engineering with AI?
Scott predicts that in the future, AI will democratize software creation by allowing anyone to describe what they want to build in natural language. This will shift the focus from implementation to ideation, enabling rapid iteration and more personalized software solutions.
What distinguishes an agent-based system like Devin from traditional AI tools?
Agent-based systems like Devin are designed to make decisions and interact with the real world, unlike traditional AI tools that provide text completion or simple answers. Devin can test its own code, run commands, and navigate through a series of tasks, making it a powerful co-worker.
What challenges does Cognition face in building Devin?
Cognition faces challenges in adapting model capabilities to real-world software engineering tasks, integrating with various developer tools, and maintaining machine state for efficient parallel work. The team focuses on solving these infrastructure and decision-making problems to enhance Devin's effectiveness.
What is the kindest thing that anyone has done for Scott?
The kindest thing Scott has experienced is the mentorship and guidance he received, especially from a mentor who pushed him to excel in competitive programming and math. This mentor invested significant time and effort, providing critical feedback and support that shaped Scott's career and success.
- Cognition built Devin, the first fully autonomous software engineer.
- Devin handles entire software workflows, including bug fixing and pull requests.
- Cognition's agentic approach to AI offers a 10x improvement in efficiency compared to traditional IDE assistants.
Shownotes Transcript