ego(lite) articles
Everything you need to know about browser automation
Guides, insights, comparisons, and reviews across Playwright MCP, Browser Use, Chrome DevTools MCP, coding agents, and the browsers they drive.
Comparisons
Playwright MCP vs CLI: token cost, setup, and trade-offs
Official workflow guidance, fixed-version CLI and MCP runs, a false-success fixture, and a clear boundary around historical token claims.
Sep 28, 2026 · 18 min read
Guides
Why computer use isn't working: debugging guide
Start with the error you can see. Check permissions, app approvals, helper sessions, and window attachment, then retry one small step.
Sep 22, 2026 · 11 min read
Guides
Fixing scraper breakage when websites change
A scraper can return zero rows and still report success. Four page versions show how extraction layers and field checks catch that failure.
Sep 22, 2026 · 12 min read
Guides
Why AI agents can't click or type in your browser
If ChatGPT can describe the page but will not click or type, treat it as a control handshake, not a broken Chrome install. Check the native host, profile, and site allowlist, then prove a click on a public page.
Sep 21, 2026 · 10 min read
Guides
Set up a remote browser workspace with Codex and Docker
Codex Remote is a phone talking to your Mac. Docker Sandboxes run Codex in a sandbox. Playwright Docker is the browser you can view over noVNC. Assemble them on purpose.
Sep 21, 2026 · 11 min read
Guides
Browser automation that survives UI changes
Durable browser automation needs more than stable selectors. The right choice depends on whether the workflow is a maintained regression path or changing live work.
Sep 21, 2026 · 13 min read
Guides
Persistent browser sessions across agent runs
A session that survives one run and dies on the next is the most common agent failure that looks like an application bug. Here is the full lifecycle: how to establish, preserve, verify, isolate, expire, and recover a browser session, with the profile-lock and session-cookie traps named.
Sep 21, 2026 · 11 min read
Guides
Context engineering for browser agents: what an observation costs
Almost every article on this term uses text-only examples. Across the top 24 results, 19 fetchable pages and 64,225 words, "accessibility tree", "screenshot" and "browser automation" each appear zero times. That is the gap this fills.
Sep 20, 2026 · 12 min read
Guides
How to test MCP server quality before deployment: tools and steps
Ship an MCP server only after the Inspector proves it connects, exposes the tools you expect, returns portable schemas, and fails the build when any of that breaks. The layers, the commands, and the CI gate.
Sep 19, 2026 · 14 min read