![cover of episode Scott Wu - Building Cognition - [Invest Like the Best, EP.402]](https://megaphone.imgix.net/podcasts/b296237a-bc20-11ef-b2ab-f38e0030a7f6/image/11b00ee2710a50b2cbdad07b056f673c.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.402]
Invest Like the Best with Patrick O'Shaughnessy
Deep Dive
What is Cognition and what does it do?
Cognition is an applied AI lab that has created the first AI software engineer, called Devin. Devin functions at the level of a junior software engineer and can handle complete engineering workflows, from bug fixing to submitting pull requests.
How sophisticated is Devin currently, and how is this measured?
Devin is currently at the level of a junior software engineer. When Cognition started, Devin was comparable to a high school CS student. Six months later, it was like an intern. Today, it's an entry-level junior engineer. Its capabilities are measured by its ability to handle tasks such as debugging, testing, and submitting pull requests.
What are some unexpected use cases of Devin that Cognition has observed?
One unexpected use case is in software migrations and modernization. Many older companies with decades-old codebases are using Devin to update their technology stacks, making their software faster, more efficient, and more secure.
What is the key difference between IDE assistants and Devin?
IDE assistants typically offer synchronous autocomplete features that speed up coding by 10-20%. Devin, however, operates asynchronously, handling complete tasks and workflows, which can save 90% of an engineer's time. Devin can test code, run commands, and integrate with various tools like Slack and GitHub.
What is the historical evolution of computer programming according to Scott?
Programming has evolved from punch cards and gas tubes to assembly, BASIC, C, and modern languages like Python. The core goal has always been to tell the computer what to do, but the form factor and complexity have changed over time, making programming more abstract and easier for humans to manage.
What qualities distinguish a 10x software engineer?
A 10x software engineer deeply understands abstractions, can solve complex problems, and efficiently translates customer needs into technical solutions. They focus on the creative and strategic aspects of software, which typically account for only 10% of an engineer's time, while the rest is spent on debugging, implementation, and other routine tasks.
What is Scott's prediction for AI in competitive programming?
Scott predicts that AI will surpass the world's best competitive programmer, Gennady Korotkevich, within one to two years. This will mark a significant milestone, similar to AlphaGo's victory in Go.
How does Cognition approach the democratization of software creation?
Cognition aims to make software creation accessible to everyone by enabling non-programmers to describe what they want to build in natural language, and Devin will build it efficiently. This shift will likely happen within five to ten years, transforming software development.
What is the impact of agents on software engineering?
Agents like Devin can handle iterative, real-world tasks, such as debugging and setting up databases. They can take a first pass at tasks, allowing humans to focus on higher-level creative work and decision-making, which significantly increases productivity.
What are the technical challenges in building Devin?
The main challenges include enhancing model capabilities for step-by-step decision-making, integrating with various tools and workflows, and maintaining a consistent state for the AI. Cognition focuses on solving these problems to make Devin more effective in real-world software engineering.
Why does Cognition charge $500 per month for Devin, and how does this compare to other AI models?
Cognition charges $500 per month because Devin performs end-to-end tasks, which are much more complex and resource-intensive than simple text completions. The price reflects the value of saving 90% of an engineer's time on routine tasks and the costs associated with running multiple Devins in parallel.
What is the competitive landscape for Cognition?
The landscape is greenfield, with the technology being very early in its development. Software is a vast and complex field, and there are many niches and use cases that can support multiple businesses. Scott believes that the focus on specific, high-impact use cases will drive success.
What is the most significant lesson Scott learned from building Lunch Club?
The most significant lesson is that it can be easier to solve a bigger problem than a smaller one. Solving a big, exciting problem helps attract a passionate and dedicated team. This was evident when building Cognition, where most of the team members are experienced founders but chose to join because of the vision and impact.
What is the defining aspect of Cognition's team dynamic?
The team has a strong foundation of trust and friendship, which makes working together and having tough conversations easier. Most team members have known each other for years, and this camaraderie is a key factor in their success and ability to move quickly.
What is the kindest thing anyone has ever done for Scott?
The kindest thing was the mentorship he received from various individuals, especially during his competitive programming days. One mentor, in particular, invested a significant amount of time and effort into pushing Scott to excel, providing guidance and harsh criticism when necessary.
- Cognition built Devin, the first autonomous AI software engineer.
- Devin handles complete software engineering workflows.
- Devin's asynchronous approach provides a 10x improvement over IDE assistants.
Shownotes Transcript