Hardware Oct 9, 2026
MacBook Air M4 long-term review: 19 months later
Nineteen months with the MacBook Air M4: where it still holds up, the one macOS bug that outlived the honeymoon, and what I would buy if it died tomorrow.
Testing AI tools on real work
I work in content and data at an internet company. This site records what happens when I point AI coding and agent tools at real tasks — not feature lists, but the debrief after a specific job either worked or fell apart.
Latest 14 entries
Hands-on Oct 9, 2026
Redesigning an Astro blog with an AI agent
Three rendered directions built from my real posts, one argument I overruled, and two environment bugs the agent caught that had nothing to do with design.
Hands-on Oct 8, 2026
Why every coding agent ships an importer for its rivals
Coding agents ship importers that name their rivals in a flag; a Google product routes to Anthropic. If the model were the moat, none of this makes sense.
Hands-on Oct 8, 2026
What the changelog says about retiring Gemini CLI
Launch posts describe intent; changelogs describe what shipped. What agy changelog revealed about retiring Gemini CLI, and why bug fixes say more.
Comparison Oct 8, 2026
Sessions in Claude Code, Muse Code and Antigravity
One treats a session as a place to return to, with seven doors back. One as a record you can audit and redact. One as the last conversation.
Guide Oct 7, 2026
Installing Gemini CLI, Codex CLI, Muse Code and Antigravity
Four installs, each under a minute. Then: Gemini CLI could not sign in, Codex CLI wanted a phone number, two binaries are not named after their packages.
Hands-on Oct 7, 2026
Muse Code bundles 20 skills, and 4 move you off Claude Code
Meta ships twenty skills with its terminal coding agent. Four exist only to migrate you off Claude Code or Codex. I pointed the importer at my own setup.
Guide Oct 7, 2026
Why an agent skill can validate and still never run
Missing frontmatter is a warning. Missing description is a warning. Only broken YAML fails. A clean validation says little about whether your skill fires.
Comparison Oct 7, 2026
Claude Code vs Muse Code vs Antigravity, compared
Bundled skills: twenty, zero, and none but a marketplace. Permissions, MCP and sandboxing read straight off each CLI — and what it does not tell you.
Comparison Oct 7, 2026
Where Claude Code, Muse Code and Antigravity keep config
One layers user, project and local settings on disk. One prints a sha256 of your policy state. One only sets your PATH. Three config models, side by side.
Comparison Oct 7, 2026
How Claude Code, Muse Code and Antigravity handle plugins
Claude Code runs plugins against a no-plugin baseline and reports the delta. Muse Code approves them capability by capability. Two questions, three designs.
Hands-on Oct 1, 2026
An SEO audit that checks Google's rules, not folklore
Four evidence tiers, one exists to say: Google never said that. Plus the two policies affiliate sites fail by accident: thin affiliation and scaled content abuse.
Hands-on Sep 30, 2026
Four places secrets leak before you publish
Git history is permanent and public. Commit authorship is readable. Screenshot pixels carry text and image metadata carries more. A scanner misses all four.
Guide Sep 27, 2026
Cloudflare Pages blog setup in 2026: eight things that broke
A static Astro site and a working inbox in one afternoon, plus eight problems no tutorial mentions: Email Routing, workers.dev taking down www, Error 1016.