ViqusViqus
Navigate
Company
Blog
About Us
Contact
System Status
Enter Viqus Hub

New Playwright Command Enables AI Agents to Record Detailed Web Demo Videos

shot-scraper video Playwright coding agents GPT-5.5 web automation CI
June 30, 2026
Source: Simon Willison
Viqus Verdict Logo Viqus Verdict Logo 7
Infrastructure Upgrade for Agent Validation
Media Hype 5/10
Real Impact 7/10

Article Summary

The post details the introduction of `shot-scraper video`, a command in the `shot-scraper` toolchain that leverages Playwright's new capabilities to record highly controlled video demonstrations. This feature requires a `storyboard.yml` file, which defines a step-by-step routine (e.g., clicking buttons, filling forms) against a running web application demo server. The author highlights that the complex YAML storyboard itself was constructed entirely by an advanced coding agent (GPT-5.5 xhigh), illustrating a powerful new workflow for demonstrating code functionality. This workflow streamlines the process of proving agent capability by providing visual evidence of complex interactions, from bulk data imports to creating tables from pasted data.

Key Points

  • The `shot-scraper video` command accepts a YAML storyboard, enabling the automated recording of complex, scripted user interaction flows.
  • The primary value of the feature is enabling AI coding agents to produce detailed, reliable video demos of their work, bridging the gap between code generation and proof-of-concept demonstration.
  • The process involves advanced tooling (Playwright, YAML definition, coding agents) that defines a robust pattern for documenting agentic workflows, making the documentation process itself highly automated.

Why It Matters

This is significant for the future of agentic systems and automated testing. Traditionally, showing an agent's work requires screenshots or abstract outputs. By formalizing the process of creating video demos via a structured YAML storyboard, the authors are providing a critical piece of infrastructure for the testing, validation, and showcase of complex, multi-step AI agent workflows. It doesn't change the core AI models, but rather improves the *interface* and *trust* surrounding the AI's output, making agent reliability a verifiable, consumable product feature.

You might also be interested in