Building agentic systems with Mrinal Wadhwa, CTO, Autonomy

"Building agents is more like engineering a live chat system or video call than it is like coding a SaaS dashboard. It’s constant, streaming, bi-directional, and that changes everything for developers." — Mrinal Wadhwa

Agents don’t behave like the SaaS apps we’ve built for decades. They’re long-running, non-deterministic, and streaming by default — closer to building WhatsApp than a web form. In this episode, Mrinal Wadhwa (CTO, Autonomy) unpacks what that shift means for engineers. From the challenges of stateful, concurrent systems to the need for new trust and authorization models, we explore how agentic software changes the rules of architecture, scaling, and reliability.

You’ll learn:

• Why traditional request–response models fall short for agent communication
• The hidden complexity of state, concurrency, and trust when agents collaborate
• Why building reliable agents is harder, not easier — even with AI-generated code
• The role of observability, evals, and guardrails in taking agents from POC to production
• How Autonomy is rethinking the runtime for multi-tenant, distributed agent fleets