Generative models for code have rapidly evolved from simple autocomplete tools into autonomous software engineering agents capable of navigating complex repositories and synthesizing multi-file patches. However, as these models take on more ambitious tasks, the fundamental bottleneck is no longer generating code – it is rigorous validation, end-to-end verification, and testing.

In this talk, we explore how agent-first development platforms like Google Antigravity shift the paradigm from manual test authoring to autonomous software verification. Rather than relying solely on static analysis or brittle, hardcoded UI scripts, we discuss how autonomous agents can be orchestrated to continuously test and validate software applications against dynamic real-world environments.

We provide an overview of how execution-based feedback, automated test generation, and browser-in-the-loop agents integrate directly into development workflows. More specifically, we describe recent advances in autonomous visual testing – where sandboxed browser subagents interact with live user interfaces, capture visual artifacts (such as video recordings and state screenshots), and reason about functional and UI regressions just as a human QA engineer would. Finally, we examine how verifiable agent deliverables and automated exploration are closing the loop between generative code synthesis and industrial-grade software reliability.