My first 24 hours with Siri AI on the Mac
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Siri AI in the macOS 27 ("Golden Gate") developer beta can launch apps but cannot take actions inside them, and AI-generated Shortcuts also failed to run Geekbench or Cinebench benchmarks automatically — one shortcut even added a "wait for you to run the test" step.
- Siri AI successfully averaged benchmark scores from screenshots selected in Finder, distinguishing single-core, multicore, and GPU results, but got confused when synthetic and time-based test screenshots were mixed and sometimes pulled the wrong data entirely.
- Siri AI is far more capable inside Apple's ecosystem — it surfaced cat and baby photos from Photos and Messages but missed Signal messages, Google Photos, and the reviewer's local Lightroom Classic catalog stored in the Pictures folder.
- Siri's Visual Intelligence, compared to last year's Copilot Vision, couldn't see all data in a Google Sheets spreadsheet until it was downloaded as an Excel file — and then returned multicore data when asked for the highest single-core Geekbench score.
- Siri AI in Lightroom Classic offered specific exposure and contrast adjustments to mimic street photographer Alan Schaller's look, but then went sycophantic, telling the reviewer they'd achieved an "almost timeless feel" — the type of behavior Apple says it won't show.
- macOS 27 remains in early developer beta with no visible indexing-progress indicator in settings; when asked, Siri told the reviewer to click a Settings button that didn't exist.
Why it matters: Siri AI's Mac-specific gaps — third-party messaging apps, non-Apple photo libraries, and Visual Intelligence's screen-view limits — mean power users who bounce between ecosystems gain only marginal workflow help, while Apple's own anti-sycophancy guidance is already being violated in its first public beta.
Ask SkimNews
