Topic
Techniques and tools for programmatically controlling web browsers, letting software or AI agents navigate and act on websites.
Cloudflare's Agents SDK browser keeps tabs, cookies and logins only until a default 10-minute idle timeout ends the session. Then it starts a fresh browser and reports restarted: true, and agent code has to check that flag before acting on a remembered login.
Reality
Hermes Agent v0.21.2 fills passwords from 1Password, Bitwarden or its own vault into web pages without the model ever seeing them. Every card fill waits for a person, so the agent cannot charge a stored card on its own, cron jobs included.
Reality
Microsoft Research's Webwright lifted GPT-5.4 from 33.5% to 60.1% on 200 long web tasks by having it write browsing code from a terminal. Because the baseline steered by screen coordinates, the result compares writing code with clicking pixels.
Reality
One dev.to author's Playwright gate has a fast model answer three typed questions per browser action, then lets plain code return allow, ask or block. The labelled evaluation is still unfinished, so the gate's unsafe-allow rate is unmeasured.
Reality
Muse passed 2.5 million downloads in its first two weeks. Its flights book through a contracted Duffel API, and its hotels book through a sandboxed browser session that navigates Expedia and Hotels.com.
Reality
An end-to-end read of the browser-automation CLI found --pin-tab written to an fsynced sidecar file and re-read at startup, while --allowed-domains is seeded only from an environment variable that a reconnecting command never sets.
Reality
Browser-use's perception pass strips scripts and hidden nodes, then paints numbered badges on a screenshot so the model clicks by index instead of by XPath. The teardown puts action precision above 95% per action.
Reality
ProgramDistill turns 26 working web applications into thousands of verifiable tasks where the specification is the running software itself. Difficulty comes from stacking repairs that depend on one another, and the reported scores track it.
Reality
Plain requests, a spoofed Chrome TLS fingerprint and headless Chrome all failed on mobile.de, and the first failure came back as 200 OK. Driving an ordinary Chrome over the DevTools protocol returned 64 listings.
Reality
An element with overflow: hidden still has a live scrollTop, so click() moved a clipped blog form 1200 pixels while the wheel and the End key moved it zero. All the suite asserted was the click.
Reality
Cowork sessions now start from desktop, web, mobile and a Chrome side panel and run on Anthropic's servers with files tied to the user's account, so whether the desktop app is open decides what the agent can read on a machine.
Reality
A Rust daemon and a Chromium extension open a separate window inside the developer's daily browser profile, so the agent inherits every live login. The documented approval prompt covers tabs you already have open.
Reality
A startup CEO told Business Insider the signups arrived in a week or two with no launch and no marketing. Cloudflare estimates bot traffic already exceeds human traffic and could reach 1,000 times it within five years.
Reality
Four Playwright jobs on one Mac held com.google.Chrome, so a Dock launch came back with error -600 while every job log stayed clean. The repair also had to survive npm install, which keeps resetting the browser path inside node_modules.
Reality
Apple's new Safari MCP server gives Claude Code or Codex the DOM, screenshots and console output of whatever the browser window is rendering. It stays off until a developer turns on two separate settings.
Reality
The open-source tool routes agent commands through a local daemon into an extension running inside your real browser profile, so every login that profile holds is in reach of whatever agent calls it.
Reality
Mininglamp published NavEval scores for its own model on its own benchmark. Across the three entries, the spread tracks how each stack reads a page. It is not a case of specialists beating frontier models.
Reality
A developer driving a real logged-in Chrome over raw DevTools Protocol logged five failures that never raised an exception, and the only thing that caught the failed upload was reading the file input back in the page.
Reality
A dev.to write-up puts agent-browser inside a UI evaluation harness so a coding agent hunts for the Settings dialog through an accessibility snapshot, with a one-line Playwright assertion deciding pass or fail.
Reality
A dev.to walkthrough holds the prompt fixed and rebuilds the code around GPT-3.5 Turbo six times until the agent really votes. Its first version exits the loop on a field the model itself sets.
Reality
Build · September 11, 2026 · One report1 publisher
Build · September 11, 2026 · One report1 publisher
Build · September 10, 2026 · One report1 publisher
Build · September 10, 2026 · One report1 publisher
Build · September 7, 2026 · One report1 publisher
Build · August 29, 2026 · One report1 publisher
Build · August 25, 2026 · One report1 publisher