Anthropic's product managers on the research team spec out requirements for each new AI model, defining what it should be good at (e.g., coding, knowledge work). This product development discipline is applied to the inherently unpredictable process of "growing" a model, bridging the gap between research and user needs.
To handle the "fire hose" of user feedback, Anthropic's PMs use Claude itself. The AI clusters feedback, identifies top themes, and even generates synthetic data based on user problems. This dogfooding creates a powerful feedback loop, turning qualitative data into actionable insights for model improvement.
Inspired by human dreaming as a memory reconsolidation process, Anthropic has its AI agents use downtime to "dream." During this background process, the agent reviews its memories, identifies and prunes contradictions, and cleans up the information to improve the coherence and utility of its long-term memory.
AI has dramatically reduced development cost, turning many engineering decisions into "two-way doors." Unlike irreversible strategic choices, code can be built and changed quickly. PMs should thus focus deep thinking on truly irreversible "one-way door" decisions, rather than on engineering time estimates for reversible work.
AI has compressed development cycles from weeks to days, but it hasn't equally accelerated human coordination. The new bottleneck is getting stakeholders aligned on strategy, planning user communication, and managing the "fuzzy" aspects of a launch. While coding saw a 100x speed-up, these coordination problems remain.
Previously, PMs needing data on feature usage filed a request and waited days. Now, they ask Claude—which has access to production databases and Slack—and get answers in minutes. This self-serve data access removes a major bottleneck, enabling faster, more fluid strategic thinking and decision-making.
Comprehensive model evaluation doesn't always require thousands of test cases. To diagnose a specific issue, like an image recognition failure, a focused set of just dozens of examples can be sufficient. This smaller, targeted approach is enough to prove a hypothesis and create a clear evaluation metric for researchers to iterate against.
Anthropic's emphasis on written communication—long-form essays, detailed docs, and in-doc discussions—creates a vast, high-quality dataset of the company's internal knowledge. This corpus serves as a powerful context source for Claude, making it more effective for internal tasks. Organizations should prioritize writing to build their own internal data advantage.