How We Built Claudie, Our AI Project Manager
A tutorial by Natalia Quintero & Dan Shipper. Featured in the OTF curated resource library.
The Vision for Claudie
Claudie was born from a simple observation: project management tools generate a lot of busywork. The team at Every asked — what if an AI could handle the administrative overhead of project management while humans focused on the creative and strategic work?
The video explores how this vision evolved from a prototype into a production tool that the team uses daily.
Technical Architecture
The architecture combines Claude's language understanding with structured project data to create an agent that can parse meeting notes, generate task lists, identify blockers, and suggest priority changes.
Key technical decisions include using a retrieval-augmented approach for project context and a structured output format that integrates with existing project management tools.
Lessons Learned
Start with a narrow scope
The team started with just meeting note summarization before expanding to full project management capabilities.
Iterate on the prompt, not the architecture
Most improvements came from refining prompts rather than changing the underlying system design.
Trust but verify
AI-generated task lists always go through a human review step before being committed to the project board.
