Tuesday • 1hr 13min

Scott Wu - Building Cognition - [Invest Like the Best, EP.401]

Invest Like the Best with Patrick O'Shaughnessy

Scott Wu is the co-founder and CEO of Cognition, an applied AI lab that has created Devin, the first AI software engineer. In just one year since founding, Devin has reached the capability level of a junior software engineer. Scott, a former competitive programming champion, brings deep expertise in both technical and business aspects of AI and software development. The conversation explores the past, present and future of software engineering, the role of AI agents, and the implications of autonomous coding systems.

0:00

-0:00

Key Takeaways

Devin is the first fully autonomous AI software engineer that functions at the level of a junior engineer, capable of handling complete engineering workflows
AI will likely surpass the world's best competitive programmer within 1-2 years, marking a major milestone in AI capabilities
The future of programming will shift from implementation to ideation - engineers will spend more time on creative problem-solving rather than routine coding tasks
AI agents represent a paradigm shift from simple text completion to autonomous systems that can make decisions and interact with the real world
The bottleneck in software development has been supply, not demand - AI tools like Devin could help unlock 10x more software creation

Introduction

Scott Wu is the co-founder and CEO of Cognition, an applied AI lab that has created Devin, the first AI software engineer. In just one year since founding, Devin has reached the capability level of a junior software engineer. Scott, a former competitive programming champion, brings deep expertise in both technical and business aspects of AI and software development. The conversation explores the past, present and future of software engineering, the role of AI agents, and the implications of autonomous coding systems.

Topics Discussed

Devin's Current Capabilities (6:29)

Scott describes Devin's current capabilities as equivalent to a junior software engineer, having evolved from a "high school CS student" level six months ago to an "entry-level engineer" today. Key aspects include:

Works asynchronously through Slack, GitHub and other standard development tools
Handles end-to-end workflows including bug fixing, testing, and submitting pull requests
Functions best for routine tasks rather than complex architectural decisions
Enables parallel work with multiple Devin instances tackling different tasks simultaneously

Early Use Cases and Customer Adoption (8:43)

Scott shares insights on how early customers are using Devin, with some surprising use cases emerging:

Code modernization and migration projects for legacy systems
Version upgrades and platform transitions for large codebases
Customers report 8-12x efficiency gains compared to manual engineering work
Focus on routine but time-consuming tasks that free up human engineers

Evolution of Programming (12:40)

Scott provides historical context on the evolution of programming:

Programming has always been about "telling computers what to do"
Progress marked by increasing levels of abstraction from punch cards to modern languages
Each major advance required finding practical applications to prove value
AI represents the next frontier in human-computer interfaces

Competitive Programming Background (16:58)

Scott discusses his experience winning the International Olympiad in Informatics:

Competition focused on algorithmic problem-solving under time pressure
Success requires deep understanding of abstractions and creative thinking
Skills transfer to real-world engineering through problem decomposition
Predicts AI will surpass human champions within 1-2 years

Future of Software Engineering (17:46)

Scott shares his vision for the future of software development:

Natural language will become primary programming interface
Custom software will become accessible to non-programmers
Focus will shift from implementation to ideation
Timeline estimate of 5-10 years for major transformation

Understanding AI Agents (23:10)

Scott explains the significance of AI agents versus simple language models:

Agents can make autonomous decisions and interact with real-world systems
Move beyond text completion to full workflow automation
Enable iterative problem-solving through multiple steps
Represent next evolution in AI applications across industries

Technical Architecture of Devin (25:45)

Scott describes how Devin is built and operates:

Integrates with existing development tools and workflows
Learns and adapts to specific codebases over time
Makes sequential decisions to solve complex problems
Builds on foundation models with specialized capabilities

Impact on Software Engineering Jobs (33:33)

Scott discusses implications for human engineers:

AI will handle routine implementation tasks (currently 90% of work)
Engineers will focus more on creative problem-solving (currently 10%)
Overall software creation capacity will increase 10x
Demand for software continues to exceed supply

Business Model and Pricing (41:44)

Scott explains Cognition's approach to pricing and business model:

Base plan at $500/month for engineering teams
Usage-based billing with "Agent Compute Units"
Pricing aimed at 10x cost savings versus traditional engineering
Value-based pricing model aligned with customer outcomes

Thoughts on AGI (43:31)

Scott shares his perspective on artificial general intelligence:

Practical impact matters more than theoretical benchmarks
Focus on real-world applications and value creation
Capabilities will improve incrementally rather than suddenly
Distribution and adoption key challenges beyond raw capability

Building the Company (54:49)

Scott discusses the startup journey and company culture:

Team of 20 people, many former founders
Strong emphasis on trust and friendship among leadership
Living and working together in shared house
Focus on speed and iteration in product development

Future Vision (1:02:27)

Scott shares his longer-term vision:

Democratizing software creation for non-programmers
Enabling "creative mode" versus "survival mode" in development
Improving idea iteration speed through rapid prototyping
Broad impact across industries beyond software

Conclusion

The conversation with Scott Wu reveals a compelling vision for the future of software development powered by AI agents. Through Devin and similar technologies, we may be approaching a fundamental shift in how software is created - from a specialized technical discipline to a more accessible creative endeavor. While challenges remain around adoption and integration, the potential impact on productivity and innovation is enormous. The next few years will likely see rapid advancement in AI coding capabilities, with significant implications for developers, businesses, and society at large.