The Personal Assistant Landscape, Right Now

Four major personal assistant launches happened in the last seven weeks, and I wanted to actually map the space instead of reacting to the headlines one at a time. Grok Bot in August. Meta's Muse in early September. Instinct raising a billion dollars at a ten billion dollar valuation on September 28th. OpenAI's Dots the very next day. Add Faff, Fo, Poke, and Town, and there are eight real products worth knowing in detail, not just by name.

the eight, side by side

Assistant	Company	Launched	Interface	Autonomy
Grok Bot	xAI	Aug 2026	Dedicated app	Full
Muse	Meta	Sep 2026	App, WhatsApp	Full
Instinct	Spear Street	Invite-only	SMS, WhatsApp	Full
Dots	OpenAI	Sep 2026	ChatGPT, Slack, Teams	Full
Faff	Faff Technologies	2024	WhatsApp	AI plus human
Fo	Wajo	Sep 2026	Own phone, inbox	Escalates to human
Poke	Cognition	Mar 2026	Texting apps	Full
Town	Town.com	Jun 2026	Web app, Gmail, Slack	Drafts, opt-in autonomy

what each one actually does

Grok Bot gives each user a persistent agent with its own cloud computer, memory, and tools, and cross-bot coordination for more complex work. It expanded to enterprises with access and audit controls on September 3rd, and by September 22nd, Tesla had shipped it into vehicles, letting drivers manage email, calendars, files, and chats by voice. Pricing starts at $120 a seat a month for teams, $200 a month for individuals.

Muse books appointments, buys tickets, fills out forms, and shops, running in its own secure browser environment with connections to Google Workspace, Ticketmaster, OpenTable, and Stripe. It's proactive and long-running, you give it a goal, it builds a plan, and it keeps working after you close the app, coming back when something changes or it needs a decision from you. It's rolled out in the US first, free or on $20 and $100 monthly tiers, and has already been downloaded millions of times.

Instinct understands the everyday texture of a life, planning a road trip, ordering weekly groceries, cancelling a forgotten subscription, booking a restaurant, all from a few words over text or WhatsApp. There's no app. It's invite-only, and it's developed a phone-concierge capability for the things that still require an actual call, doctor's appointments, cancellations, high-end bookings. The billion-dollar round on September 28th, led by Sequoia, Benchmark, and Coatue, values it at ten billion dollars.

Dots, announced the day after Instinct's raise, runs on GPT-6 Astra and connects to more than 4,000 apps through OpenAI's plugin ecosystem. It's explicitly meant to be hardware and interface agnostic, pursuing a goal continuously with minimal oversight, and OpenAI's own framing is a future where a user runs a small team of Dots working together, not just one assistant.

Faff lives entirely in WhatsApp and is built for the non-work admin that keeps getting pushed to tomorrow, groceries, bookings, repairs, travel, gifting. It pairs AI with real people on the other end, and it's completed more than 75,000 tasks since its 2024 founding in Bengaluru, on $2.84 million raised from Nexus Venture Partners and Infinyte Club.

Fo takes human-in-the-loop in a different direction than Faff does. The agent has its own phone number, its own inbox, and its own payment card, and it emails the landlord and pays for the booking directly wherever the task is digital and structured. But when the job is an actual phone call to a business with no app and no API, most plumbers and barbershops, a real Wajo staffer makes that call, not a synthetic voice. Wajo's own framing is blunt about it, other personal AIs pretend AI can do everything, Fo employs humans to do the parts AI genuinely can't. Wajo claims Fo is twice as good at real-world task completion as competing agents and holds a 94 percent trust rate, and opened it to the public on September 28th after a month of invite-only alpha.

Poke is fully autonomous, no human fallback, but it never asks you to leave the messaging app you already have open, iMessage, WhatsApp, Telegram, RCS. It handles daily planning, calendar management, fitness tracking, smart home control, and photo editing, and talks back with more personality and slang than any of the others. It exchanged over 100 million messages in a single quarter, became the first AI agent approved for Apple's Business Messages platform, and Cognition, the company behind the coding agent Devin, bought it outright in July.

Town gives each user a "Townie," an agent that builds a working model of how they operate from their email, documents, and patterns, then drafts, schedules, researches, and runs recurring routines like morning briefings and inbox triage across more than 50 work tools, Gmail, Slack, HubSpot, Notion, Linear, GitHub. Nothing sends without approval by default, replies land in your drafts folder in your voice and wait for a yes, and more autonomy is something you opt into per routine, not something the product assumes going in. It raised a 55 million dollar Series A led by Andreessen Horowitz on June 3rd, alongside Forerunner, First Round, Alt Capital, and Conviction, the same day it came out of beta.

the gap nobody's filled yet

The empty spot is a hybrid with real product depth, a human-in-the-loop assistant that also plugs directly into booking and payment rails the way Muse does, instead of routing everything through a chat thread. Faff and Fo both prove the trust model works. Neither has built the Ticketmaster-and-OpenTable-level integration that would make the same model feel as fast as the full-autonomy players. Town takes a similar ask-first stance in its own lane, nothing sends without approval by default, but its 50-plus integrations are work tools, not the consumer commerce and booking rails this gap still needs. There's a second gap sitting in plain sight too, every one of these eight is a US or India company. Nothing here is built for Southeast Asia, Latin America, or Europe specifically, and the behavioral assumptions baked into a WhatsApp-first product or an SMS-first one won't transfer cleanly to a market with different messaging habits.

where this consolidates

Cognition buying Poke in July already answered the question of whether this space consolidates, the only open question is who's next and why. The most likely acquirers are the full-autonomy giants who eventually run into the same liability wall I wrote about months ago, and the fastest way to buy an honest answer to it is to acquire a team that's already built one, which makes Faff and Fo both plausible targets, not just Poke's category of texting-native distribution. The most likely sellers are whichever of the smaller players start running out of runway before their model gets proof at scale, and given the gap between Instinct's ten billion dollar valuation and Faff's under-three-million-dollar raise, that runway question hits very differently depending on which side of this map a company is sitting on.

what changes in how people behave

The detail that actually unsettles me about Fo isn't that it might be impersonating a human on that call, it's that it quietly is one. When the plumber picks up, there's a real Wajo staffer on the line, not a synthetic voice, routed invisibly through a product you think of as your AI assistant. That's a different behavioral shift than the one I expected, not "can you tell it's a bot," but "do you even know which parts of your assistant are AI and which parts are a person getting paid to pretend the boundary doesn't exist." Poke leaning into slang and personality is a related but different bet, make the agent feel more like a person on purpose, and Town is the inverse of both, draft everything, send nothing without a yes, which only makes sense once you stop assuming the system on the other end is fully either one.

what's actually the same across all eight

Every one of them is chasing the same underlying shift, from an assistant you query to an assistant you assign a goal to. None of these are search bars with a personality anymore, all eight plan multiple steps, hold a task across time, and come back to you only when something needs a decision. All eight also lean on existing infrastructure rather than rebuilding it, calendars, payment rails, booking platforms, work tools, someone else's messaging app, instead of a walled garden. And every single one is still figuring out monetization in public, free tiers, usage tiers, or in Grok Bot and Fo's case, pricing high enough that the company is betting real value gets delivered, not just novelty.

where they actually differ

The sharpest line used to be autonomy versus human fallback mapping neatly onto capital, the cautious players small, the confident ones well-funded. Town breaks that. Faff and Fo both bet a person should stay in the loop, and both are among the smallest and least capitalized of the eight. Town bets the same way, nothing sends without approval by default, but it just raised 55 million dollars from Andreessen Horowitz, the first well-capitalized player here to choose caution over full autonomy on purpose rather than by necessity. The second line is interface, Grok Bot, Muse, Dots, and Town all built or lean on a dedicated app or web product, while Instinct, Faff, Fo, and Poke live entirely inside text and messaging, no download required. The third is geography, Faff is the only one built for and operating out of India, everything else here is a US company, even where the underlying model serves a global user base.

who's actually ahead on what

Distribution, Poke and Instinct, neither makes you download anything or learn a new interface, they show up inside a text thread you already have open. Depth of integration, Muse, its direct ties into Ticketmaster, OpenTable, and Stripe make specific transactions cleaner than a general-purpose agent guessing its way through a checkout flow. Hardware reach, Grok Bot, shipping inside an actual car dashboard is a different level of ambient than living on a phone. Honesty about limits, Fo, building the product around recognizing its own failure points is a more mature answer to trust than promising full autonomy and hoping it holds. Capital efficiency, Faff, 75,000 completed tasks on under three million dollars raised is a real, working business, not just a demo. Ambition, Dots, the explicit plan is a team of agents working together, not one assistant, and that's a further-out bet than anyone else here is making publicly yet. Trust design, Town, the only well-capitalized player that chose draft-and-approve by default instead of full autonomy, proof that caution and real money aren't actually mutually exclusive.

conclusion

Eight products this fall, and only one real fault line underneath all of them, does the assistant ask first, or does it ask forgiveness. Full autonomy moves faster and breaks more. Human-in-the-loop moves slower and breaks less. Nobody's run either model long enough to say which one a person actually wants living inside their calendar and their card.

Four launches in seven weeks wasn't a coincidence, it was a category deciding its own shape in real time. I don't think it slows down before the next one arrives, and I don't think the next one looks like any of these eight.