A Verge reviewer spent a day with the rebuilt Siri in Apple's macOS 27 developer beta and found a voice assistant that finally answers questions well, but still cannot act inside apps.
A 24-hour test of Apple's rebuilt voice assistant on a MacBook Pro makes one thing clear: Siri AI in the macOS 27 developer beta is finally good at answering questions, but on a machine built for keyboards and mice, answering questions was never the hard part.
That is the verdict from Antonio G. Di Benedetto, who covers Mac laptops for The Verge and spent just over a day with the new assistant in Apple's "Golden Gate" developer beta of its next Mac operating system, writing in a hands-on published Friday. His central thesis lands inside the first scroll: "Siri is better, but its limitations are much more obvious on a Mac than an iPhone." That tension, between a smarter assistant and a surface that punishes the friction voice input adds, is the story worth sitting with.
The improvement is real. Di Benedetto, who had turned Siri off on the Mac years ago and never re-engaged, ran the new build through a full day of real work and found the assistant measurably more useful on factual lookups, on summarizing what was on his screen, and on the conversational follow-ups that used to collapse into dead ends. That is the part of the review that matters most to anyone considering re-enabling Siri in the beta: the basic information-retrieval layer is finally competitive.
What it still cannot do is the part the Mac makes a person feel. The new Siri, like its predecessor, cannot reliably take actions inside applications. Asking it to rename a layer in Photoshop, refile a Finder selection, or run a multi-step automation across apps still routes the requester back to the keyboard. On an iPhone, where touch is the bottleneck and the screen hides what the assistant is doing, that limitation is forgivable. On a Mac, where a keyboard shortcut is already the fastest path to the same outcome, an assistant that can only talk at the screen about what is on it is a step sideways, not forward.
Di Benedetto's review is filed inside The Verge's WWDC 2026 coverage hub, and the reviewer's track record matters here. He has covered Mac laptops at The Verge since 2021 and arrived after 15 years in professional photography, so he knows the keyboard-first workflow the Mac is actually designed for. He also comes to the beta with a stated bias: he had previously dismissed Apple Intelligence, Apple's earlier AI effort, as "fruitless." That prior is worth naming, because it means the positive observations have to clear a higher bar to land.
Two questions will determine whether the beta matures into a shipping feature worth installing. The first is indexing: the assistant needs to know what is on the Mac to answer questions about it, and the state of that index is opaque in the developer build. The second is in-app action: whether Apple ships a permissions and intents layer that lets the new assistant actually do things inside third-party apps, not just look at them. The first question is engineering. The second is harder, because it depends on developers opting in.
For now, the practical takeaway is the one Di Benedetto's 24 hours actually earned. Siri AI in the macOS 27 developer beta is good enough at answering questions that turning it back on is a reasonable experiment. It is not yet good enough at acting inside the apps where Mac work actually happens, and on this surface that gap matters more than it does on a phone. Re-enable it for the questions, keep the keyboard close for the rest, and watch the indexing and in-app action story as the beta hardens toward a public release.