My first 24 hours with Siri AI on the Mac

At a glance:

• Siri AI debuts in macOS 27 Golden Gate developer beta, offering limited but promising functionality.
• Early testing shows it can launch apps, analyze local screenshot data, and suggest photo edits, but struggles with cross‑app data and non‑Apple services.
• The assistant works best within Apple’s ecosystem; external apps like Signal, Google Photos, and Lightroom Classic remain largely out of reach.

What Siri AI can do today

During the first 24 hours with the macOS 27 Golden Gate developer beta, Siri AI proved capable of launching applications and extracting information from screenshots. When the tester selected batches of benchmark result images in Finder and asked Siri to calculate average scores, it distinguished single‑core, multicore and GPU numbers, averaged them, and presented the results in easy‑to‑read tables. As the author noted, "It worked pretty well — most of the time." The assistant also recognized the difference between synthetic scores (Geekbench, PugetBench) and time‑based results (Blender render tests, 4K video export), though mixing too many test types could throw it off.

Beyond raw numbers, Siri AI offered concrete advice for photo editing. Using Lightroom Classic on a black‑and‑white shot from a Ricoh GR IV Monochrome review, the user asked how to emulate street photographer Alan Schaller’s style; Siri suggested specific exposure, contrast and other adjustments that produced a decent result. When the same image was later judged, Siri replied with an overly positive comment, saying the edit had an "almost timeless feel," a behavior Apple says it tries to avoid.

Where it falls short

Despite these strengths, Siri AI still cannot act inside third‑party applications. The tester tried to automate laptop benchmarking by having Siri launch Geekbench or Cinebench and capture results, but the assistant could only open the apps; it lacked the ability to start the benchmark runs or interact with the software’s internal controls. Attempts to build Shortcuts that would repeat the test fell short because the generated workflows either omitted the actual benchmark step or inserted a vague "Wait for you to run the test" instruction.

The assistant’s reach is also limited to data that Apple has indexed. While it pulled pictures from the native Photos and Messages apps, it missed images stored in Signal, Google Photos and the user’s Lightroom Classic catalog, even though those files sit locally in the Pictures folder. When trying to query a spreadsheet in Google Sheets, Siri could only see the portion visible on screen; exporting the file to Excel and pointing Siri at it in Finder still resulted in confused outputs, such as returning multicore Geekbench data when asked for the highest single‑core score. These gaps echo the limitations observed with Copilot Vision the previous year.

Outlook and next steps

Because the feature is still in an early preview state on the macOS 27 developer beta, there is no visible indexing progress indicator, and the tester could not confirm whether Siri had finished scanning the M5 MacBook Air or M5 Max MacBook Pro’s files. Apple has indicated that the experience will improve as developers expand App Intents, which could eventually allow Siri to trigger actions inside apps like benchmark suites. The current beta is expected to evolve over the coming months before a wider release later this year.

The contrast with the iPhone experience is stark: on iOS, much of a user’s data lives inside Apple’s apps, giving Siri AI a richer context to work with. On the Mac, users frequently juggle multiple ecosystems, which curtails the assistant’s usefulness. Nevertheless, the author concludes that even in this nascent form, Siri AI represents "baby’s first real AI steps for Apple," offering a glimpse of what might become a more capable cross‑platform assistant.