Look at what the first wave of consumer AI went after - a lot of founders went after sexy stuff - AI companions, journaling bots and mental-wellness apps: soft products with no transaction anywhere in them. These were essentially a prompting machine and we all know what happened to most (the only ones that flourished were..soft p0rn companions).
The Personal AI space of 2026 is totally opposite - transactional, boring and measurable.
Personal AI agents win on the chores people hate: cancelling subscriptions, booking travel, chasing a refund.
Instinct, the texting agent that raised $1 billion at a $10 billion valuation in late September 2026, grew on exactly those errands.
Meta is making the same bet with Muse, wrapped in a much bigger pitch.
My take is that the winners in this category will be the teams that pick one dull, recurring job, finish it every time, and earn trust one permission at a time. In fact, the success will be proportional to boringness of the task.
Instinct and Muse: Race for super intelligence
In the last few weeks, two very different people pitched the same idea.
Mark Zuckerberg went first. In a WSJ op-ed and an essay on Meta’s site, he argued for “personal superintelligence”: AI that sits with individuals instead of being concentrated in a few institutions. When Meta launched its Muse agent on 8 September, the promise was agents working “24/7 on your behalf to improve your relationships, health, career, finances, home management, hobbies, and more.”
Noah Shinn’s pitch is smaller, and I think sharper. Shinn is the 23-year-old founder of Instinct, a Northeastern dropout who worked at Sierra and co-authored the Reflexion paper on language agents. On Patrick O’Shaughnessy’s Invest Like the Best podcast in late September, his product rule came through clearly: focus on understandability first, capabilities later. Users should always know what the agent is about to do. His users text Instinct to cancel subscriptions, plan weddings and book trips.
Travel alone is about half of the roughly $1 billion in annualised transaction volume the company reports.
Both describe the same category, but Zuckerberg is selling a future and Shinn is selling a finished errand. I’d back the errand, anyday!
What Jobs-to-Be-Done says about personal AI
Clayton Christensen’s Jobs-to-Be-Done idea is simple: people don’t buy products, they “hire” them to get a job done. The famous example is the morning milkshake, hired by commuters to make a boring drive less boring and keep them full till lunch. Nobody was buying “dairy innovation”.
Most AI products so far have sold capability. A bigger context window, a better benchmark score, a smarter model. JTBD says nobody pays for capability on its own; they pay for the outcome.
Personal AI is the first AI category where the product is the outcome. People will hire an agent because their gym makes them cancel in person, or because rebooking a cancelled flight means waiting on hold with an airline. Those jobs share a few properties that suit agents well:
A clear “done” state. The subscription is cancelled or it isn’t. A chatbot answer has no such finish line, so you can never quite tell whether it worked.
They repeat. Bills and subscriptions come round every month, and frequency builds the habit that investors are paying $10 billion for.
Everyone hates them. Irritation is a stronger buying trigger than curiosity; people will try an unknown product to escape a customer-care IVR.
A mistake is cheap. A botched cancellation costs you a follow-up message, which makes these the right jobs for a new agent to learn on.
This is why I call personal AI the biggest JTBD bet in tech right now. Every other AI category can hide behind demos. Here the user finds out the same day whether the job got done.
Why starting with chores is the right call
Look at what the first wave of consumer AI went after. A lot of founders built companions, journaling bots and mental-wellness apps: soft products with no transaction anywhere in them. The best a user could say was “this AI gets me”, and there’s no way to measure that, price it or check it the next morning. Without a done state, nobody could tell whether the product was working (only soft porn worked :D) , including the founders.
Personal AI is the opposite. It is transactional and boring, and every task ends in an outcome you can verify: the refund landed, the flight is booked, the subscription stopped renewing. That is exactly why it works as a business.
Founders still love the sexy version of personal AI: the companion that knows you, the chief of staff that runs your life. It demos well on stage. The trouble is that nobody hands their life to software on day one.
Instinct’s own numbers show how trust actually builds. Shinn told O’Shaughnessy that about 40% of users share credit card details after three weeks, which means the card comes after a run of smaller jobs.
The agent that cancels your streaming plan in week one is the one you let book your holiday in week four. Chores are the trust ladder, and every rung is a low-risk job done right.
The chores also drive distribution, because a line like “it cancelled my gym membership while I was in a meeting” gets repeated at dinner. Fortune reports Instinct has spent $0 on marketing, runs invite-only with five invites per user, and that invites have been resold on eBay. On the podcast Shinn put the growth this way: “Every day, about 10% of the audience... are making a decision to give up one of their five valuable invites to somebody else.” The company has not disclosed user numbers, so treat that as the founder’s figure.
“No app” is the product decision
The design choice I find most interesting in Instinct is the one it left out - Instinct has no app in the usual sense. You text it or call it, and it can call you back; it has its own phone number and works through SMS, iMessage and WhatsApp. Shinn’s longer view, from the same conversation: “I think that all of software is going to collapse down into... honestly a single, very, very easy to use interface.”
Poke, from The Interaction Company, made the same choice on iMessage and crossed 100 million messages in the three months before Cognition bought it on 23 July 2026. Meta put Muse in its own app and on WhatsApp.
The JTBD reading: the job happens where the user already is. Nobody wants to open a new app to get rid of a chore; they want to fire off a message the way they would to a dependable assistant.
For India, that place is WhatsApp, full stop.
The more important fight is happening on the merchant side. On 20 September Amazon blocked Muse from shopping on Amazon.com, citing lack of authorisation and how Muse stores customer credentials. Tech Times pointed out that Amazon’s own Buy for Me agent shops on third-party sites without asking them first.
A week later, on 28 September, Shopify partnered with Instinct. The real battle in personal AI is over who controls the final “buy” button, and the platforms with ad businesses to protect will not hand it over quietly.
Where the open opportunities in India are
Instinct and Meta are fighting for the general-purpose agent. That leaves a lot of ground for founders who want a specific job to own. Here is where I’d look.
1. Vertical agents for the jobs people dread most. In India the list writes itself: passport and visa paperwork, health insurance claims, telecom and broadband disputes, property tax, chasing an e-commerce refund. Each has a clear done state, deep irritation, and process knowledge a general agent won’t have. A narrow agent that closes insurance claims faster than a human agent would be a business on its own - though my belief is that most of these will involve a lot of human intervention, so very debatable on whether such companies can ever be AI-native (unless we want to believe that AI chatbots make one..)
2. A WhatsApp-native personal agent for India. Muse’s WhatsApp access launched in the US only. India has the habit (WhatsApp) and the payment rail (UPI), where every payment already needs the user’s PIN. That per-transaction approval is close to how Instinct asks users to approve one-time Stripe Link cards for each purchase. The trust primitive already exists here; someone needs to build the agent on top of it.
3. Agent-ready commerce for merchants. Shopify’s deal with Instinct tells you where this goes. Every D2C brand and marketplace will need a way to accept orders from agents: structured catalogues, agent checkout, identity checks. Most Indian D2C brands are nowhere near ready. Tooling for them is a clear B2B opportunity.
4. The trust and liability layer. Instinct’s terms of service, as reviewed by eesel, authorise the agent to act as your legal representative and leave you liable for actions you didn’t intend. Somebody will build the audit logs, dispute handling and possibly insurance that make users comfortable delegating more. Permissions are currently each company’s own system (Sentinel at Meta, the Vault at Instinct); a neutral layer has room.
5. Agents for families. Instinct already runs a “trusted person network” where agents coordinate plans with other users’ agents. Now think of an Indian professional in Bengaluru or Toronto managing their parents’ electricity bills, doctor appointments and bank KYC from a distance. That is a recurring, high-anxiety job, and nobody owns it.
6. The receiving end. When millions of personal agents start calling clinics, salons and restaurants to book slots, someone has to answer. Small businesses will need their own agents to handle agent traffic. Zuckerberg argues personal superintelligence will help people start businesses without much capital; the first tool those businesses need may be an agent that talks to customers’ agents.
What’s your take?