😸 Meta reads your mind

gm legends, happy Tuesday.

Today: an email assistant that does the admin you hate without making you switch apps, a browser agent that refuses to click anything, and Meta reading full sentences straight out of people's heads. Plus the forum on whether to pick a crowded market or a brand-new one.

Supafax came out of Rohan Mahendraker watching his girlfriend lose hours to email and calendar admin for three years, so he and Anton Muratov (YC W26) built it to do that admin for her: an assistant that lives inside your existing Gmail or Outlook, handles the busywork on its own, and only emails you when it needs a decision. No new app, no new client, no prompting, and it never stores your email.

🔥 Our Take: Rohan watched his girlfriend lose hours a day to email admin for three years, so he built her something to handle it.The clever part is what it doesn't make you do: no new app, no new email client, nothing to learn. It sits inside your normal Gmail or Outlook, quietly does the boring admin, and only pings you when it needs you to decide something. Every other "AI for your email" wants you on their shiny new inbox, and nobody wants to move house just to answer emails. The thing I'd keep an eye on is Google and Microsoft putting Gemini and Copilot right inside Gmail and Outlook, so Supafax has to stay clearly better than the free assistant your inbox is about to ship with. But doing the work where your email already lives, with nothing to set up, is the most sensible version of this I've seen.

Pluno is a browser agent that skips the clicking entirely, talking straight to the APIs underneath your web apps instead of loading pages and pressing buttons like a human, so it can update records, pull data, or run bulk SaaS chores in the background. Built by the Abstreiter brothers off the back of an earlier AI support startup, it benchmarked at 34% more accurate and about 14x faster than Claude's browser extension across 312 tasks in 24 tools.

🔥 Our Take: You know that thing where you watch a browser agent do a task and it's weirdly slow and painful? It loads the page, screenshots it, stares at it, clicks the wrong button, gets stuck in a dropdown. Pluno's whole point is the agent's doing it the human way for no reason, when it could just talk to the API under the app directly, like code does. So instead of clicking around HubSpot or Notion it learns the API and hits it straight, which they clocked at 34% more accurate and about 14x faster than Claude's browser extension. And it fits the founder, Korbinian, an ETH data-science guy who spent years building AI for customer support, so "stop clicking, just call the API" is exactly the instinct you'd expect. The catch nobody really mentions is that learning some app's private API is fragile, the apps never agreed to it, and it can't get past a CAPTCHA.

ElevenLabs has been the go-to for voice for a while. Now they've turned that expertise into agents that actually get things done. You set one up, it talks like a real person, listens, responds, and helps handle the task — support calls, bookings, whatever the job is. Not a demo, not a "press 1 for sales" situation. It's ready to deploy. Feels like one of those shifts where the interface quietly changes. Less typing, less clicking, more just saying what needs to happen and letting it play out.

Brain2Qwerty v2 is Meta's research system for typing with your mind: it decodes full sentences from non-invasive brain recordings, hitting 61% word accuracy on average and 78% for its best participant, trained end-to-end on raw brain signals with the code released openly. The reader doing the listening is an MEG scanner the size of a room.

🔥 Our Take: Meta got a machine to read whole sentences straight out of someone's head, no surgery, nothing implanted, and it's right about 61% of the time (78% for their best person). It's a big jump because they let the AI learn straight off the raw brain signals instead of the hand-built pipelines people used before, then open-sourced the lot. The thing keeping it in the lab is the reader: an MEG scanner, a room-sized machine that isn't going anywhere. So it's not Neuralink, nobody's drilling into your skull, but the price of skipping surgery is a magnet the size of a fridge. A real milestone for the no-surgery kind of mind-reading, and also miles from your desk. The accuracy's climbing fast. It's the hardware that has to shrink, and that's the hard part.

Aleksandar Blazhev (@byalexai) asked the question every founder circles at some point: do you go after an established market with proven demand and a crowd of competitors, or carve out a brand-new niche nobody's named yet?

Most landed in the same place: established market, but with a razor-sharp angle. Anna Ludwinowski (@anna_ludwinowski) argued the founders winning either way aren't differentiating on market maturity at all, they're differentiating on specificity, being crystal clear about one person's exact problem everyone else treats as generic. A few told on themselves too, including one founder whose mental-health app was "neither unique enough to carve a new niche nor differentiated enough to compete in an established one," the classic middle-ground death.

Mustafa (@thenameisarian) reframed the whole thing: problem age and format age are separate variables. New niche means both are new, which is brutally hard. The winning move is usually an old problem in a new format, "most successful category creation was actually format creation on an old problem." Figma on design, Superhuman on email, Substack on writing.