Zach Wright Email Zach
All Updates

Here is what you missed while you were shipping.

The Big Thing

The best agent web stacks now decide whether to read, extract, or browse before they touch the page.

Why it matters: too many workflows still treat every web task like browser automation. That is getting expensive. Clean markdown paths, cheap fetch APIs, and search-first extraction tools now let operators reserve full browser sessions for the few jobs that actually need state, JavaScript, auth, or clicks.

Code & Tools

  1. Vercel content negotiation - serve markdown from the same blog, changelog, or docs URL and give agents a lighter, cleaner read path. https://vercel.com/blog/making-agent-friendly-pages-with-content-negotiation https://vercel.com/changelog/docs-pages-support-markdown-responses
  2. Browserbase Fetch API - read the page without launching Chromium, then decide whether the task actually deserves a browser. https://www.browserbase.com/blog/fetch-api
  3. Firecrawl Skill & CLI - one install gives coding agents scrape, search, crawl, map, and browser commands that write results to the filesystem instead of bloating context. https://www.firecrawl.dev/blog/introducing-firecrawl-skill-and-cli https://www.firecrawl.dev/blog/firecrawl-official-claude-plugin
  4. Exa Deep - query expansion plus grounded summaries improves source selection before you fetch or crawl anything downstream. https://exa.ai/docs/changelog/new-deep-search-type
  5. Sandboxed browser escalation - Firecrawl Browser Sandbox and Browserbase Functions give you isolated execution when auth flows, forms, or pagination force a real browser. https://www.firecrawl.dev/blog/introducing-browser-sandbox https://www.browserbase.com/changelog/browserbase-functions

Tech Impact

Meme of the Day

"Tech Loops" (xkcd) - because the fastest way to fetch one page is apparently to benchmark three web-agent stacks first.

Image URL: https://imgs.xkcd.com/comics/tech_loops.png
Post: https://xkcd.com/1579/