Building in public: insights on agentic collaboration, AI-human teamwork, and the tools we're creating.
AI keeps making production cheaper. The less obvious change is that as output accelerates, coordination, review, and decision-making become more expensive. The scarce resource is no longer execution—it's shared understanding.
Read more →Code generation got cheap. Alignment didn't. We built a 16-task benchmark to measure where durable project memory matters most—and found a 68 percentage point improvement in decision recall, context rehydration, and cross-thread synthesis.
Read more →Both teams produced good plans—until one question required verifying the past. The difference wasn't intelligence. It was whether the team could cheaply check institutional memory.
Read more →Watercooler is a structured memory system for humans and agents, designed to preserve team context, reasoning, decisions, and critique over time.
Read more →A literal water cooler conversation changed how we think about creativity. Watercooler is our attempt to engineer that effect back into human–AI collaboration.
Read more →Watercooler has reached a baseline where it's useful, not just aspirational. We're starting a build-in-public series to produce a new 2-minute demo—and replace the one on our landing page.
Read more →