In this video
What you'll learn
The vibes problem
Why "it passed our tests" isn't enough; the PM/Legal/Ops trust gap (origin story).
The mental model
What an eval actually is: input → expected behavior → judge. The 4 kinds that matter: correctness, quality, compliance
Hands-on
Catch agent failures before your users (or your customers) do. Leave with a reusable framework.
Why this topic matters
Building an AI agent is easy now. Trusting it isn't. Most teams ship on vibes, then the
first edge case breaks everything. The skill that fixes it,writing evals, is barely taught.
In 60 minutes, I'll walk you through it with field notes and real failures from shipping
agents to production.
You'll learn from
KD Deshpande
AI Founder, Building the Future of AI Agents & Agentic Workflows.
KD Deshpande is a 2x founder, former product leader at Meta, Uber, and Adobe, and the founder of Simplified, an AI Native platform serving 15M+ users. Over the last several years, he has worked hands-on with GPTs, AI agents, and multi-agent systems in production, helping companies turn AI from a demo into real business impact. In this session, he'll share practical lessons from building and scaling AI-native products used by millions,
Previously at
Keep exploring
