Blog
Notes
Route-first web agents: benchmarks, comparisons with browser automation tools, and guides to getting website data without scraping. RSS
2026-10-04 · Opinion
In Defense of Muse’s Most Uncomfortable IdeaA useful personal agent needs delegated account access. The serious question is how to bound that access and make the resulting work accountable.
2026-10-04 · Opinion
The Second Run Is the Test of Your Browser AgentA browser agent should reuse what it has learned. Test changed inputs and honest failure before calling a recorded request a reusable operation.
2026-10-04 · Opinion
The Agentic Web Will Not WaitSearch indexed what the web knows. Unbrowse indexes what the web can do. Agents should not have to wait for every website to wrap its capabilities again.
2026-10-04 · Opinion
The Invoice Is Right ThereAn assistant should absorb the work around a task. Count what still falls to the customer before celebrating another connected account.
2026-09-28 · Opinion
We Killed the Apify Actor. Agents Build Them Now, for FreeAn Apify actor is a scraper a person wrote and maintains. Unbrowse has an agent compile the same thing from a site's own requests, free in beta.
2026-04-02 · Opinion
Browser Automation Is Dead. Here's What Replaces It.Browser-first agent work often pays for rendering and page parsing when the structured route already exists underneath. Across 94 domains in the paper benchmark, warmed cached routes averaged a 3.6x mean speedup over Playwright.
2026-04-02 · Opinion
MCP Wrappers Are Not a Coverage StrategyA common tool interface makes capabilities easier to call. It cannot supply an operation the integration never implemented.
2026-04-02 · Opinion
Agents Need an Index They Can Act OnA useful capability index helps an agent choose an operation, understand its access requirements and judge whether the result is trustworthy.