Boris Cherny, Head of Claude Code, shares how AI tools like Co-Work can autonomously manage complex tasks such as booking multiple flights and hotels by intelligently analyzing user data, marking a significant leap in AI capabilities and user interaction paradigms. He emphasizes that while AI adoption is rapidly advancing and driving productivity gains, organizations must experiment and adapt their workflows and culture to effectively integrate AI, rather than relying on superficial usage metrics or isolated cases of misuse.
Boris Cherny, Head of Claude Code, shares his experience using an AI tool called Co-Work to autonomously book multiple flights and hotels for his upcoming travel schedule. He describes how the AI reviewed his emails and calendar, identified missing stops and incorrect dates, and successfully booked eight flights and five hotels with minimal intervention. Cherny highlights that this level of autonomous task completion is a significant improvement compared to previous AI capabilities, emphasizing the rapid monthly advancements in AI technology that require users to continuously update their expectations and experiment with new models.
Cherny discusses the paradigm shift AI agents represent in how people interact with technology. Unlike traditional software with fixed interfaces and features, AI agents can personalize and shape online experiences based on user preferences, effectively acting as autonomous assistants. This shift is driving explosive growth in AI adoption, as users increasingly rely on AI to perform complex tasks on their behalf, moving beyond manual interactions with software.
Addressing concerns about “token maxing”—the practice of artificially inflating AI usage metrics to meet corporate targets—Cherny expresses skepticism that this is a widespread issue. Drawing from his experience at Facebook and Anthropic, he explains that genuine productivity gains from AI are substantial, with some companies reporting hundreds of percentage points increase in code output per engineer. He advocates for giving employees freedom and psychological safety to experiment with AI tools, as innovation often comes from unexpected individuals across various roles, not just top engineers.
Cherny acknowledges reports of token maxing at companies like Amazon but suggests these may be isolated cases rather than the norm. He emphasizes that the real challenge for organizations is how to integrate AI into their business processes effectively. Drawing a parallel to the adoption of personal computers in the 1990s, he notes that productivity gains only materialize when companies restructure their workflows around new technology rather than treating it as a peripheral tool.
In conclusion, Cherny underscores that the adoption of AI is still in an experimental phase for many organizations, with no one-size-fits-all solution. Companies are exploring different strategies to harness AI’s productivity benefits, and success depends on cultural and organizational factors. He encourages a mindset of continuous experimentation and adaptation to fully realize the transformative potential of AI in the workplace.
