Beyond Coding

← Beyond Coding19 aug · 42 min

Amazon AI Lead: What Differentiates The Best AI Coding Models

Amazon AI Lead: What Differentiates The Best AI Coding Models19 aug42 min

<p>How does Amazon build its agentic AI? Michael Giannangeli, Head of Product for Amazon Nova and Agentic AI, breaks down evals, RL gyms, and model routing. </p><p>He also explains why the bottleneck in software has shifted away from engineering hours and what takes its place.</p><p>In this video, we cover:</p><ul><li>The eval lifecycle: building from real failure modes, saturation, and why 100% means delete</li><li>RL gyms: training models on real environments like migrations, DevOps, and pen testing</li><li>Model routing, cost-per-token trade-offs, and why routing isn't solved</li><li>The agent stack of an Amazon product lead: Claude Code, Codex, and Kiro</li><li>Autonomous migrations, trust, and how much human-in-the-loop survives</li></ul><p>For engineers and product people building with AI agents who want to see how a frontier lab actually closes its feedback loops.</p><p>Recorded at the AI4 conference 2026. </p><p><br /></p><p>Timestamps:<br />00:00:00 - Intro<br />00:00:36 - The Agents an Amazon Product Lead Uses Daily<br />00:03:36 - Why Nobody's Heard of Amazon Nova<br />00:04:55 - Model Costs and the Routing Problem<br />00:08:10 - Why Building Good Evals Is So Hard<br />00:10:05 - When Evals Saturate and Get Deleted<br />00:12:17 - Turning Real Failure Modes Into Hundreds of Evals<br />00:15:26 - Improving Models Without Training on Customer Data<br />00:18:26 - If Everyone Uses Agents, You Need Agents<br />00:20:22 - The Bottleneck Is No Longer Engineering Hours<br />00:23:20 - Ship Fast to Validate the Right Thing<br />00:26:44 - Staying at the Frontier Amid Constant Noise<br />00:29:37 - Spend 10-20% of Your Time Experimenting<br />00:32:54 - RL Gyms: How Models Learn From Failure<br />00:37:09 - Will Migrations Become Fully Autonomous?</p><p><br /></p><p>Guest: Michael Giannangeli - Head of Product, Agentic AI &amp; Amazon Nova at Amazon</p><p><a href="#AmazonNova" rel="ugc noopener noreferrer" target="_blank">#AmazonNova</a> <a href="#AgenticAI" rel="ugc noopener noreferrer" target="_blank">#AgenticAI</a> <a href="#AIEngineering" rel="ugc noopener noreferrer" target="_blank">#AIEngineering</a></p>