Driving the Agent Quality Flywheel from Your Coding Agent

2026-06-30 · Source: Google Developers Blog - AI · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Intermediate, long

Summary

A new developer-facing skill, released on June 30, 2026, drives a "Quality Flywheel" methodology for coding agents, aiming to systematically improve agent quality by integrating evaluation into the development workflow. Developed in partnership with Google DeepMind, this skill is centered on a five-stage "Build & Test" loop: Prepare Data, Run Inference, Grade, Analyze Failures, and Optimize & Iterate. It utilizes Google's adaptive AutoRaters and supports custom metrics, ensuring the optimizer and evaluator remain decoupled. The system can evaluate agents using synthetic scenarios via the User Simulator or by grading production traces. It is available in two packages, `google-agents-cli-eval` for ADK agents and `agent-platform-eval-flywheel` for any framework, requiring a GCP project with the Agent Platform GenAI Evaluation Service enabled.

Key takeaway

For AI Engineers tasked with ensuring robust agent quality, this new skill offers a disciplined approach to move beyond "vibe-checking." You should integrate the `google-agents-cli-eval` or `agent-platform-eval-flywheel` skill into your coding agent workflow. This enables systematic evaluation against production or synthetic data, isolating failures with custom rubrics, and iteratively optimizing agent behavior, ensuring changes genuinely improve performance rather than introducing regressions.

Key insights

Systematic agent quality improvement requires a disciplined evaluation flywheel with decoupled optimizers and evaluators, driven by coding agents.

Principles

• Decouple optimizer and evaluator.
• Trust score deltas over absolute grades.
• Production data refines evaluation loops.

Method

The method involves a five-stage Build & Test loop: Prepare Data, Run Inference, Grade, Analyze Failures, and Optimize & Iterate. This process uses AutoRaters and custom rubrics on synthetic or production traces for iterative quality improvement.

In practice

• Install `google-agents-cli-eval` for ADK.
• Install `agent-platform-eval-flywheel` for SDK.
• Use User Simulator for cold-start eval.

Topics

• Agent Quality
• LLM Evaluation
• Coding Agents
• Google Cloud AI Agent Platform
• AutoRaters
• OpenTelemetry Tracing

Code references

• google/adk-samples
• google/agents-cli
• google/skills

Best for: AI Engineer, Machine Learning Engineer, MLOps Engineer

Related on AIssential

• Whitepaper Companion Podcast - Agent Quality — Agent quality requires architectural integration, focusing on the entire execution trajectory, and continuous evaluation with robust observability.
• Evals Skills for Coding Agents — Improving infrastructure around AI agents, especially evaluation capabilities, is more critical than solely improving the underlying model.
• I Stopped Coding and Started Architecting Agents (And You Should Too) — Harness engineering uses guides and sensors to constrain generative AI agents, creating a flywheel effect for continuous code quality improvement.
• Stop Shipping on Vibes — How to Build Real Evals for Coding Agents — Robust AI evaluations are crucial for quantifiable shipping decisions, performance monitoring, and iterative improvement.
• TestSprite launches an open-source command-line tool to help AI agents check their own work — AI coding agents can self-verify their work using a new open-source CLI tool, improving code quality and reducing hidden bugs.
• Demystifying evals for AI agents — Effective AI agent evaluation requires combining diverse grading techniques and structured methodologies to match agent complexity.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Google Developers Blog - AI.