Building Predictable Conversational AI

Write your awesome label here.

Yam Marcovitz unpacks what it takes to build predictable, production-ready AI. From human-in-the-loop alignment to test-driven development, we explore how Parlant helps teams design reliable, flexible, and transparent AI systems.

In this episode of Unprompted, we sit down with Yam Marcovitz, former software architect at Microsoft and Apex.AI, and co-founder of the open-source AI framework Parlant. With a background in real-time systems and enterprise software, Yam brings a systems-level perspective to one of AI's biggest challenges: getting LLMs to consistently do what you want.

We explore the gap between design and engineering in conversational AI, why current tools fall short, and how open source plays a crucial role in responsible AI development. Yam shares practical insights on building agent behavior that's not just intelligent but reliable and predictable at scale.

This episode covers:

Why alignment is a human problem, not just an algorithmic one: How team collaboration and clear communication impact AI behavior as much as the technology itself
The limitations of current conversation design tools:What's missing from today's approaches and why most teams struggle with consistency
How Parlant separates concerns: Empowering both designers and engineers to work effectively without stepping on each other's toes
Creating flexible agents without sacrificing control: Maintaining brand safety and predictable behavior while allowing for natural conversation flow
Building test-driven development for conversational AI: Moving beyond "prompt and pray" to systematic testing and validation
Managing incidents and updates efficiently: How to fix problems and deploy changes without retraining models or starting from scratch
Designing for multilingual, multichannel interactions: Scaling conversation design across languages and platforms
How linguistic precision shapes agent behavior: Why the way you describe actions and responses directly impacts performance
The case for transparency and open infrastructure: Why open source is essential for building AI systems that can scale responsibly

Whether you're deploying conversational AI at enterprise scale or just trying to get your bot to behave consistently, this episode offers a sharp, systems-level look at the future of conversation design.

Formerly a software architect at Microsoft and Apex.AI, Yam brings a unique combination of deep technical expertise and systems thinking to AI development. With extensive experience in high-performance computing and system architecture, including building realtime systems with microsecond-level precision for autonomous vehicles, Yam understands what it takes to create reliable, production-grade software. This background informs his distinctive approach to AI development in Parlant—an open-source framework that fundamentally reimagines how we control AI agent behavior.

Drawing from his experience in both low-level systems optimization and high-level architectural design, Parlant introduces a structured system for guiding AI behavior that emphasizes flexibility, precision, reliability, and practical development workflows, reflecting Yam's commitment to building robust, production-ready systems.