Masonry: Okay, this is the kind of update that makes me think, all right, voice finally stopped being a demo toy. If I’m already inside the app and I can just steer work with my voice, that’s a real workflow collapse.
Eyre: Yeah, and the important part is that it’s not just speech-to-text with a shinier coat. GPT-Live is doing the listening and speaking at the same time, so the system can keep state while you’re still talking through what you want.
Masonry: Right, and that matters because the user story is not “please admire my voice quality.” It’s, I’m in three threads, I’ve got a thing running in Work or Codex, and I want to poke it without doing the whole tab-dance.
Eyre: Exactly. The desktop app update says you can start a new chat or task in voice mode, then ask ChatGPT to start, check, or steer work in other threads. That’s the first time this feels like an actual control surface, not a side feature.
Masonry: And that is so very Exploring Next of them, by the way. We get suspiciously excited any time someone turns a dropdown into an operating system for a human afternoon.
Eyre: Mm-hm.
Masonry: The Mac bit is the sneaky one to me. Appshots can look at the active window in focus, so the voice model has a better shot at understanding what you mean without you narrating your screen like a court deposition.
Eyre: That is the part with the clean mechanism. It’s still bounded context, but it’s context from the actual foreground window, which is much more useful than hoping the user describes the right file name out loud.
Masonry: And then they’ve got it working with Computer Use, local files, and plugins too. So this isn’t voice as a separate island. It’s voice as the front door into the same messy pile of tools people are already trying to use.
Eyre: Right, and the plan gating tells you who they think this is for. Plus, Pro, Business, Edu, Enterprise on desktop. That’s not casual consumer novelty territory. That’s people who already live in the app long enough to care about shaving friction.
Masonry: I also think the multi-folder project thing is quietly nice. You can set a primary folder, but keep secondary folders around for search, reading, and editing. That’s just less annoying than forcing one folder to pretend it contains the whole shape of a project.
Eyre: And the primary folder still drives new chats, Git operations, and the automatic discovery of AGENTS.md, skills, and config.toml. So they’re preserving a canonical root while letting the rest of the workspace breathe a little.
Masonry: Which is the right compromise, honestly. Too many tools act like the project is a single neat box, and then everybody’s surprised when real work has three subtrees and one weird folder nobody wants to rename.
Masonry: Also, the native Mac feel complaint is fair. I get why people roll their eyes at the app getting bigger while it learns new tricks. But if the trick is actually useful, people forgive a lot faster than product teams think.
Eyre: Sure, but I’d separate useful from inevitable. The clever part here is the coordination layer. The questionable part is the app sprawl, because now ChatGPT, Work, Codex, voice, and file context are all living in one surface and that can get noisy fast.
Masonry: Yeah, but noisy surfaces are what real adoption looks like. Clean little toy apps don’t get used for work. Messy tools with one sharp affordance do.
Eyre: Oh, that’s annoyingly right.
Masonry: And if you want the zoom-out, this is basically OpenAI saying the desktop app is the workspace, not just the chat window. GPT-5.6, GPT-Live, Work, Codex, voice. It’s all getting pulled into one place.
Eyre: That’s the broader race too. OpenAI is clearly trying to make the app itself the control plane, while everyone else is still arguing about whether voice is a feature or a modality. This is the part where the product answer matters more than the slogan.
Masonry: Honestly, I think a lot of people will bounce off the first minute and then keep using it once they realize they can just say, ‘check that thread’ instead of hunting around. I’d put sixty-forty that the voice button becomes a habit for desktop users faster than people expect.
Eyre: Mm, maybe. My hesitation is that voice only sticks when the task is already awkward enough to benefit from it. But for anything involving multiple agents, active windows, and file context, this is a pretty good shape.
Masonry: Which is a very Eyre way to say, ‘fine, it’s actually useful.’
Eyre: Yeah, don’t get cocky.
Masonry: Come on, Eyre, I’m allowed one little victory lap. We’ve been doing this since November and somehow this is where the desktop app grows up. I cannot believe this is how we’re spending a Thursday.