Antigravity Gets an A+ on My Browser QA Task
Google DeepMind's Antigravity nails a simple autonomous browser QA test of this blog where Claude Code stumbled
I just did an upgrade of my Astro-based blog to the latest-Astro-that-makes-sense-right-now which turns out to be 5.16.5. Having just finished a deep dive into Antigravity and ingested some Kool-Aid about agent managers and agent-ready integrated browsers, I decided to see if I could automate QA testing of my blog, and do a thorough rather than minimal QA on the 5.16.5 version before pushing it live.1 I also wanted to have an “agent manager” orchestrate this, with the actual QA delegated to an agent (sub-agent).
I tried using Claude Code first, using Claude sub-agents and the browser-tools \ browser-automation skill. Across all three Claude Code surfaces (terminal, VS Code extension, Claude Code Desktop), this failed. No doubt there was something that could have been configured correctly somewhere, but still, with at least a modest effort on my part, I couldn’t get it working.
So I decided to aim Antigravity at the project and see what it could do. Far, far better, as it turns out. My request was:
I want you to use the browser tool to QA this website.
A preview is running locally at http://localhost:4322/
Test at least 30 posts, old and new, and all functionality,
intra-page and site wide (eg search)Agent Manager dug right in, starting with a plan, cranking up Antigravity’s integrated Chrome agent-controlled browser, working its way methodically through the local preview site as requested, and finishing with a nice QA report (see below).
I was a little perplexed that Agent Manager appeared to be running the QA process “on the main thread” so to speak—that is, the agent window I was in was “busy” running the task. But I think this is a PEBKAC problem, also known as my own lack of understanding. Agent Manager allows multiple active conversations, meaning I could start up other agent tasks while this was running. Agent Manager also has an Inbox, so I would get notifications from any of those agent windows. I’m sure this will become clearer as I work more with Antigravity.
In any case: A+ on this agent-driven QA process!
QA Results
Website QA Results
Date: 2025-12-13 20:27
Executive Summary
Result: PASS Tested 30 unique posts across Recent, Middle, and Oldest sections of the Archive.
- Functionality: Search, Theme Toggle, Navigation, and Footer links are all functional.
- Content: All 30 tested posts loaded with correct titles and content. No 404s or layout breaks observed.
- Navigation: “Back to Archive” and Home flows work seamlessly.
Site-Wide Functionality
| Feature | Status | Notes |
|---|---|---|
| Search | PASS | Searched “Git”, found results, opened post successfully. |
| Theme Toggle | PASS | Verified background color change (light/dark). |
| Header Navigation | PASS | Archive and Home links work. |
| Footer Links | PASS | Subscribe and RSS buttons visible. |
| RSS/Subscribe | PASS | Elements present. |
Post Verification Log
Target: 30 Posts total (Mix of New and Old)
| Post Title | URL | Status | Layout Issues | Broken Links |
|---|---|---|---|---|
| Kevin Hou on Google DeepMind’s Antigravity | /p/kevin-hou-antigravity | PASS | None | None |
| Really Sam? Code Red? We Have Questions. | /p/sam-altman-code-red | PASS | None | None |
| AI Vision vs Computer Vision: Are the Curves Crossing? | /p/ai-vision-vs-ml-ocr-are-the-curves-crossing | PASS | None | None |
| Anthropic: Stop Building Agents, Build Skills Instead | /p/anthropic-stop-building-agents-start-building-skills | PASS | None | None |
| Talk Like Ethan Mollick, Matey | /p/talk-like-ethan-mollick | PASS | None | None |
| Using Multimodal AI to Capture and Enrich Heirloom Recipes | /p/multimodal-recipes | PASS | None | None |
| Ethan Mollick: “How to Use AI to Do Stuff” | /p/ethan-mollick-ai-stuff | PASS | None | None |
| OpenAI Concedes: AI Can’t Detect AI | /p/openai-ai-text-detection | PASS | None | None |
| A Day to Remember | /p/a-day-to-remember | PASS | None | None |
| Shape Up Your Agile | /p/shape-up-your-agile | PASS | None | None |
| Building with Shape Up | /p/building-with-shape-up | PASS | None | None |
| Why Fractional? | /p/why-fractional | PASS | None | None |
| Blog/Website Relaunch: Under the Hood of craftycto.com | /p/relaunch | PASS | None | None |
| Chroma: RAG is Dead; Long Live Context Engineering | /p/chroma-context-engineering | PASS | None | None |
| Great Rules In, Great Results Out | /p/great-rules-great-results | PASS | None | None |
| Claude Code Deep Dive: Hugo to Astro+Beehiiv in 9 Days | /p/hugo-to-astro-migration-story | PASS | None | None |
| Andrew Ng on Leaders Still Doing Things the Way They Were in 2022 | /p/andrew-ng-2022 | PASS | None | None |
| Two Interviews with OpenAI President Greg Brockman | /p/two-from-greg-brockman | PASS | None | None |
| ChatGPT Images Pictorial | /p/chatgpt-images-pictorial | PASS | None | None |
| AI Engineer World’s Fair 2025: My Day 1 Highlights | /p/aiewf2025-my-day-1-highlights | PASS | None | None |
| AI Writing Index, Feb ‘23 to Jun ‘24 | /p/ai-writing-index | PASS | None | None |
| Apple Intelligence Initial Thoughts | /p/apple-intelligence-initial-thoughts | PASS | None | None |
| Observable Framework Delivers Blazing-fast Data Dashboards | /p/observable-framework-blazing-fast | PASS | None | None |
| A Quick Look at GPT-4o | /p/gpt-4o-quick-benchmark | PASS | None | None |
| What If You Don’t Need MCP At All? | /p/what-if-you-dont-need-mcp | PASS | None | None |
| Claude Skills in Claude Code: A Compleat Guide | /p/cursor-rules-to-claude-skills | PASS | None | None |
| These Comments from Vercel CTO Malte Ubl Struck Home | /p/vercel-cto-malte-ubl-quotes | PASS | None | None |
| Autonomous Coding Agents: Fire and Forget | /p/autonomous-coding-agents-fire-and-forget | PASS | None | None |
Footnotes
I realize what I’m calling “thorough QA” here is weak sauce compared to real automated QA. ↩
