Did the degraded variants also end up with weaker test coverage, or was that controlled separately? Because the fix for "agent burns more tokens understanding the code" is very different from "agent gets confused by dead code" — the first needs better tests and docs, the second just needs a linter pass.
The practical bottleneck for most AI tooling isn't model capability anymore - it's the orchestration layer. Getting reliable behavior across edge cases requires more engineering than people expect from demos.
MCP for browser automation is interesting because Safari's WebKit engine is the one most AI agents can't easily drive (Playwright and Puppeteer are Chromium-first). Having an MCP server for it could fill a real gap in cross-browser testing for agent workflows.
MCP for browser automation is interesting because Safari's WebKit engine is the one most AI agents can't easily drive (Playwright and Puppeteer are Chromium-first). Having an MCP server for it could fill a real gap in cross-browser testing for agent workflows.
The hard part of Jest/Vitest isn't the API — it's understanding async behavior, mock boundaries, and when to test implementation vs. behavior. Does this cover those patterns or mostly syntax?