Interview
Cognition CEO Scott Wu on Acquiring Windsurf: The Process, The Deal, The Rationale
- Existing AI code progress alone is expected to fundamentally alter the world even without new research breakthroughs.
- AI coding tools are predicted to evolve from the current 1.5x–2x efficiency range to 10x efficiency gains within three years.
- Increased efficiency is projected to trigger a Jevons paradox, driving total code volume to increase by more than 10x.
- Software agents will eventually assume ownership of tasks, transitioning human roles to technical architects or product managers who express intent rather than write code.
- Programming is forecast to shift toward high-touch intent expression where humans interact directly with computers without viewing code.
- Deep context capabilities will become the primary research focus to address practical challenges in complex codebases and debugging.
- Agent user experiences are anticipated to achieve IDE-level familiarity within the next six to twelve months.
- The developer software market is expected to evolve into a clean, non-code interface for intent expression resembling a human-computer interface like Jarvis.
- Both IDEs and agents will remain vital components of workflows in the immediate future before merging into a combined experience.
- Strategies involve maintaining distinct philosophies for Devon and Windsurf products while engineering a smooth user experience between them.
- The market for major foundation model players is predicted to consolidate to between two and six companies, with three to five being a reasonable expectation.
- Aggregate valuations for foundation model companies such as OpenAI, Anthropic, and xAI are expected to rise significantly over the next five years.
- Aggregate valuations for application layer AI companies including Perplexity, Cursor, and Harvey are also expected to rise significantly over the next five years.
- Industry data strategies will shift from prioritizing mass data volume to utilizing small, highly curated datasets for specific use cases.
- Core skills in solving problems from first principles, understanding computer models, and grasping system architectures will remain valuable for computer science education.
- A computer science degree is recommended for new students as it teaches essential thinking skills regardless of AI advancements.
- AI coding tools will generate immense value, though the specific percentage collected by technology builders (ranging from 5% to 30%) remains uncertain.
- Value accrual is expected to occur wherever real differentiation is established, potentially favoring application layers over foundation labs alone.
- The code and software building sector is expected to have multiple winners rather than a single monopoly.
- The acquisition strategy was adjusted to incorporate a more personal narrative following a recognition that the previous personality-less approach was a missed opportunity.
- Inherited marketing teams will be leveraged to enhance selling capabilities and narrative-shaping efforts.
- The narrative surrounding agents is expected to shift from private development to public adoption as the technology is integrated more broadly.