Deployed: The AI Product Podcast What 2+ Years Building Enterprise AI Taught Me: A Conversation with Ema's CEO, Surojit Chatterjee
Oct 24, 2025
Surojit Chatterjee, CEO and founder of Ema, shares his journey building enterprise AI employees since early 2023. He emphasizes the critical difference between treating AI as employees rather than agents, impacting feedback and evaluation. Surojit discusses the failure of many AI projects due to automation of broken processes and promotes multi-agent systems as essential for managing enterprise complexity. He also highlights the importance of redesigning workflows, sustainable scalability, and effective training for business users to manage AI effectively.
AI Snips
Chapters
Transcript
Episode notes
Leverage Models, Build the Infrastructure
- Don't build another foundational model; instead, build infrastructure that leverages existing LLMs and composes tooling.
- Focus engineering on arms, legs, and control to make LLMs useful for mission‑critical enterprise tasks.
Multi‑Agent Composability Beats Monoliths
- Single monolithic agents lack composability, observability, and explainability needed for complex enterprise tasks.
- Ema uses multiple specialized agents (experts) and an auto builder to compose workflows and speed implementation.
Ship Prebuilt Agents, Then Customize
- Start with pre-built agents and AI employees, then configure and compose them for customer-specific workflows.
- Reuse and learn from prior workflows to accelerate new builds and gain network effects.