- Cognition's Devon AI merges 30-40% of enterprise pull requests, acting as a 'junior engineer'.
- CEO Scott Wu believes AGI is already present, citing AI's ability to solve real-world problems.
- The company rapidly acquired Windsurf in a weekend, highlighting the fast-paced AI M&A landscape.
Scott Wu, the visionary CEO of Cognition, is not just building an AI coding agent; he's orchestrating a fundamental shift in how software is developed. From his unconventional path as a math prodigy and Harvard dropout to leading the charge in agentic AI, Wu offers a compelling glimpse into the future of technology, where AI takes on an increasingly autonomous role.
Wu's journey into AI is rooted in a lifelong passion for math competitions, a path that unexpectedly led him to drop out of Harvard and gain early experience in the Bay Area alongside a cohort of future tech founders. This unique background informs his perspective on the 'moneyballification of everything,' where deep mathematical understanding and first-principles thinking now trump intuition in mature fields like poker, chess, and even startups. He argues that as spaces mature, they resolve to their underlying mathematical truths, making it harder for young founders without extensive experience, yet simultaneously opening new frontiers for those who can master the new 'playbook.'
At the heart of Cognition's mission is Devon, an AI software engineer designed to operate at the ticket or project level, handling tasks from bug fixes to large-scale migrations and documentation. Wu reveals that Devon is currently deployed in thousands of companies, merging an impressive 30-40% of all pull requests in successful organizations. He differentiates Devon's asynchronous, agent-based paradigm from synchronous IDE tools, emphasizing its role in tackling 'accidental complexity'—the routine, repetitive tasks that consume 80-90% of an engineer's time. This allows human engineers to focus on 'essential complexity,' making high-level strategic decisions, with Devon executing the groundwork, leading to 8-15x productivity gains in enterprise migrations.
Looking ahead, Wu makes several bold predictions. He asserts that 'we have AGI' today, challenging conventional definitions by highlighting AI's growing capability to solve complex real-world problems, even if defining the 'benchmark' remains a practical challenge. He also foresees a future, within 2-4 years, where software engineers will no longer primarily interact with code, instead instructing AI agents at a higher conceptual level, much like today's engineers don't write assembly. This shift, he believes, will lead to more software engineers, not fewer, as the demand for software continues to outpace current development capacity. The recent rapid acquisition of Windsurf, executed from Friday to Monday, further underscores Cognition's aggressive strategy to integrate synchronous IDE experiences with Devon's async agent capabilities, accelerating their vision for the future of software engineering.
“I think that there will come a point and my guess on this point is probably in the neighborhood of let's say 2 3 4 years from now where we stop using code as the main interface.”
- Scott Wu, Co-founder and CEO of Cognition
