The Agent Network — Dharmesh Shah

Key takeaways

1. Shah defines an agent simply as AI-powered software that accomplishes a goal, then sorts agents by traits like autonomous or not, deterministic or not, and chat or workflow.
2. He expects hybrid teams of people and AI agents, and built Agent.ai as a professional network where agents have profiles, connections and posts, much like human hires.
3. Shah would rather under-engineer than over-engineer: the interest on tech debt is known and payable, and if the feature never needs it you saved the time. He thinks cheap AI code generation strengthens this case.
4. He warns that when adding features becomes nearly free, teams may bolt on too many, which makes products more complicated for users and hurts the business.
5. Agent.ai has 1.3 million users, 3,000 people who have built an agent and about 1,000 published agents; Shah funds a free MCP server there that gives access to many agents and models with one key.