Idea to Product Journey at Anthropic
by Benjamin Mann on July 20, 2025
Ben Mann, co-founder of Anthropic, describes a strategic approach to innovation and product development in AI that focuses on anticipating future capabilities rather than current limitations. This approach has been instrumental in creating breakthrough products like Claude Code.
The "Skate to Where the Puck is Going" Innovation Model
-
Build for future capabilities, not current limitations
- "Don't build for today, build for six months from now, build for a year from now"
- Recognize that features working at 20% reliability today will reach 100% reliability as models improve
- Internalize the exponential improvement curve of AI capabilities
-
Create structured pathways for innovation
- Define clear "journey of an idea from prototype to product"
- Establish graduation criteria for projects to become products
- Implement sprint models that maintain effective execution
-
Focus on leverage points that maximize impact
- Identify platforms that can exist in multiple environments (e.g., terminals for Claude Code)
- Look for interfaces that can scale across different use cases
- Build on foundations that will remain relevant as technology evolves
-
Balance ambition with practicality
- Ensure teams are "working on the right ambition level of thing"
- Create operating models that support innovation while maintaining focus
- Use strong management to guide teams through uncertainty
-
Learn from historical innovation models
- Draw inspiration from successful innovation labs (Google's Area 120, Bell Labs)
- Adapt proven approaches to the unique challenges of AI development
- Recognize that "it's really hard to do right" and requires continuous iteration
-
Maintain a long-term perspective
- Understand the exponential trajectory of AI capabilities
- Consider how tasks that seem impossible today will become routine
- Build with future capabilities in mind, not just current limitations
This approach has enabled Anthropic to create products like Claude Code and Model Context Protocol by anticipating capabilities that weren't fully realized when development began.