Ex-Uber dev explains his Multi-Agent Workflow

What this video covers

Crivello argues that agent use is moving from a single-player model, where each person runs a separate assistant, to a multiplayer model built around a shared AI teammate. Teams currently pass generated files and instructions between disconnected tools, while a shared agent could participate directly in the same messages, documents, email and decisions as its human colleagues.

The foundation is automatically maintained shared context rather than a static company wiki. Meetings, messages and other work streams can continuously update a common memory, allowing an agent to retrieve prior decisions, understand current priorities and offer advice that reflects what is actually happening across the organization.

This changes the human role from manually operating every workflow to designing and improving the system that carries out the work. The discussion recommends active experimentation, acceptance of short-term messiness and close attention to how agents are adopted, while retaining human judgment for creative direction, priorities and the broader consequences of automation.