Uploaded June 2026 | Updated September 2026, 2 weeks ago
Treating agent quality as a rigorous engineering discipline is the only way to scale. Stop guessing and start measuring. Join this session for a deep dive into the state-of-the-art techniques Google uses to build our own agents, and learn how to make them a part of your process. We’ll demonstrate how you can adopt the “Quality Flywheel” methodology, which includes bootstrapping effective offline evaluation with synthetic test generation, using LLM-as-a-judge autoraters and trajectory evaluations, performing user and environment simulation, identifying systemic failures in production with multi-turn autoraters and loss-pattern clustering, aligning test coverage with actual usage, and using automated optimization capabilities to scientifically refine performance. Ramp up with confidence.
Watch more Google Cloud Next 2026 Sessions → https://goo.gle/gcloud-next-2026
Subscribe to Google Cloud Tech → https://goo.gle/GoogleCloudTech
Speakers: Dima Melnyk, Alex Martin, Daniel Lewis
BRK3-023
#GoogleCloudNext
Treating agent quality as a rigorous engineering discipline is the only way to scale. Stop guessing and start measuring. Join this session for a deep dive into the state-of-the-art techniques Google uses to build our own agents, and learn how to make them a part of your process. We’ll demonstrate how you can adopt the “Quality Flywheel” methodology, which includes bootstrapping effective offline evaluation with synthetic test generation, using LLM-as-a-judge autoraters and trajectory evaluations, performing user and environment simulation, identifying systemic failures in production with multi-turn autoraters and loss-pattern clustering, aligning test coverage with actual usage, and using automated optimization capabilities to scientifically refine performance. Ramp up with confidence.
Watch more Google Cloud Next 2026 Sessions → https://goo.gle/gcloud-next-2026
Subscribe to Google Cloud Tech → https://goo.gle/GoogleCloudTech
Speakers: Dima Melnyk, Alex Martin, Daniel Lewis
BRK3-023
#GoogleCloudNext








![Stop building slow AI: Optimizing multi-agent systems for production
Build a Multi-Agent Marathon Planner with ADK and A2A → https://goo.gle/4nyWbRE
[Demo] Race Condition → https://goo.gle/3Pbv16Z
Join Casey West and Ivan Ardini as they break down the multi-agent architecture behind their Dev Keynote demo, revealing the secrets to optimizing complex AI systems for production constraints. Discover how to leverage the Agent Development Kit (ADK) to build specialized planner, simulator, and evaluator agents, and learn advanced techniques for real time evaluation using Gemini 3.1 Pro. Explore how to optimize Agent2Agent Protocol (A2A) communication using Pub/Sub, WebSockets, and protocol buffers, and master Agent-to-User Interface (A2UI) generation to build responsive, scalable frontends for cloud applications.
Chapters:
0:00 - Intro
1:14 - Considerations for building the Cloud Next 2026 demo
3:42 - Agents as judges
5:20 - Tools vs skills for AI agents
7:55 - Evaluating AI agents
11:33 - Using Agent2Agent Protocol (A2A) in the demo
16:20 - How can developers trust the output of AI models?
18:41 - Data governance for multi-agent systems
19:50 - Getting started with codelabs
Watch more Google Cloud Next 2026 → https://goo.gle/next-talks-2026
🔔 Subscribe to Google Cloud Tech → https://goo.gle/GoogleCloudTech
#GoogleCloudNext
Speakers: Casey West, Ivan Nardini
Products Mentioned: Agent Development Kit, Gemini, Agent2Agent Protocol, Pub/Sub, WebSockets, Agent-to-User Interface, Gemini Enterprise Agent Platform Stop building slow AI: Optimizing multi-agent systems for production](https://i.ytimg.com/vi/ge5cLd8uics/mqdefault.jpg)

![No design skills? Use this free AI powered design tool from Google | The Agent Factory
Stitch → https://goo.gle/4njvHU6
Introducing “vibe design” with Stitch → https://goo.gle/49dPdMj
If youve ever shipped an app that worked perfectly under the hood but looked like it was built in 2003, this episode is for you. Designing beautiful, responsive user interfaces is notoriously difficult, but what if you could outsource the heavy lifting to an AI? In this episode, we explore Stitch, Google Labs’ AI-powered UI generation tool that acts as your personal creative director, building stunning interfaces on the fly without writing a single line of initial CSS.
Host Nohe sits down with David East, DevRel Engineer at Google Labs, for a complete walkthrough of the Stitch platform. Youll see David design a fully custom Maryland crabbing tour website from scratch, establish non-negotiable constraints like color and theme, and dive deep into Design.md—the secret file that translates your creative intent into tokenized AI values. The episode also features an incredible demo using the Gemini CLI and the Stitch MCP server to pull down production-ready HTML and Tailwind CSS straight to your local environment, followed by a rapid-fire hot takes round on the future of frontend engineering.
Whether you’re a backend developer avoiding Flexbox or a seasoned architect looking to speed up your prototyping, youll walk away from this episode knowing exactly how to prompt an AI for highly-specific UI designs. Youll learn the concept of semantic compression, how to leverage design variants, and why treating AI as your creative partner rather than just a code generator yields the best results.
Chapters:
0:00 - Introduction to David East & Stitch
1:43 - What is Stitch? The AI design agent
6:08 - Getting started in the Stitch UI
14:43 - Why Design.md is the secret sauce of AI design
21:47 - Generating design variations in Stitch
30:18 - Extracting HTML and Tailwind CSS directly
34:14 - Gemini CLI & the Stitch MCP Server
40:21 - Let AI write the complex CSS grid instructions
41:47 - Hot Takes: Is app development all agents now?
More resources:
Bad at CSS → https://goo.gle/48Ogwws
Google Labs community → https://goo.gle/48UzKk5
David East’s website → https://goo.gle/4txjQUf
Speaker: Nohe [@nohedev], David East [@_davideast]
Watch more of The Agent Factory → https://www.youtube.com/watch?v=qBOvM7SiDa4&list=PLIivdWyY5sqLXR1eSkiM5bE6pFlXC-OSs
🔔 Subscribe to Google Cloud Tech → https://goo.gle/GoogleCloudTech
Products Mentioned: Google Stitch, Gemini, Gemini CLI, Google Labs, Firebase, Firebase Studio, Jules, Tailwind CSS, HTML/CSS, Jetpack Compose, Swift UI, Flutter, React, Model Context Protocol (MCP) No design skills? Use this free AI powered design tool from Google | The Agent Factory](https://i.ytimg.com/vi/hNHE301qF1A/mqdefault.jpg)