Reading up on Playwright

Taste in software cannot be unit-tested because it relies on tacit, contextual judgment that resists full externalization into rules or code.

Simon Willison introduces shot-scraper video, a tool that records video demos from a YAML storyboard using Playwright, enabling coding agents to demonstrate their work.

LLM agents can perform manual QA testing via markdown instructions, catching regressions and psychological issues, raising software quality beyond traditional automated tests.

Sentry cut build times from 14 to under 4 minutes by replacing its headless CMS with Astro, Markdown files, and Claude Code skills for content management.

Webwright lets AI agents control browsers via terminal commands and reusable scripts, achieving 60.8% on Odysseys and 86.7% on Online-Mind2Web with only 1K lines of harness code.

Cursor's cloud agents perform best when given full development environments, durable execution via Temporal, and a harness that shifts control to the agent.

Temporary scaffolding around AI models becomes obsolete as models improve, so developers should build these harnesses cheap enough to discard.

Cloudflare renames Browser Rendering to Browser Run, adding Live View, Human in the Loop, CDP access, WebMCP support, session recordings, and 4x higher concurrency for AI agents.

Anthropic engineer Prithvi Rajasekaran builds a multi-agent harness with planner, generator, and evaluator agents that produces full-stack applications over multi-hour autonomous coding sessions.

Cloudflare launches a managed crawl endpoint that respects robots.txt, aiming to provide a well-behaved alternative to aggressive AI scrapers.

Effective agentic coding requires a robust test harness, letting AI explore ideas freely, and maintaining a scratch pad of learnings between sessions.

Simon Willison introduces Showboat and Rodney, CLI tools for coding agents to create Markdown demos and browser automation scripts to verify their work.

Providing AI agents with automated feedback loops (back pressure) enables them to self-correct on long-horizon tasks, increasing engineer leverage.

Stagehand's `agent()` function enables fully autonomous browser workflows using AI, with support for multiple agent modes, custom tools, streaming, and structured output.

The article describes a technical process extracting Fizzy's daily git screenshots and scoring a flipbook video with Claude-composed music from the app's own audio samples.

DeepSeek-OCR compresses image tokens using an encoder to reduce input size, enabling larger context windows and more efficient training, not traditional OCR.

Browserbase forked Chromium and applied patches to remove automation signals and enable long-lived sessions, making headless browsers undetectable and stable for AI agents.