What is Wajo Fo, the AI agent that hires humans?

What is Wajo Fo and why is Vinod Khosla betting on it?

Wajo Fo is a personal AI agent that runs errands for you. When software alone can't finish the job, it hands the task to a paid human assistant. Wajo, the San Francisco startup behind it, was founded by Shivani Poddar, who led engineering at Google DeepMind and worked on AI at Meta before that. Fo opened to the public on September 28, 2026, after a month as an invite-only alpha. Within days it had a famous supporter. Vinod Khosla, whose firm has backed Wajo from the start, says Fo will win the crowded agent market on one thing: trust. That's a bold claim in a field where Meta's Muse, OpenAI's Dot and Instinct already have a head start.

His reasoning is simple. An agent acts in your name, so what matters isn't how clever it is. What matters is whether you'd hand it your card and your contacts. Khosla pointed to three design choices. Every purchase goes through a new single-use virtual card, so the merchant never sees your real card details. Fo has its own phone number and email address and never sends messages as you. It can also sign in to websites for you without seeing or storing your password. He took a swipe at the competition too, asking who wants to give all their data to Meta. Whatever you think of the pitch, a top investor is selling an agent on restraint, not raw power.

Should you take the numbers at face value? Not yet. Wajo says Fo completed 71% of 104 real-world tasks in its own testing, and that it beats rival agents both at finishing tasks and at keeping private data private. Those are the company's own tests, on tasks it picked, and the headline comparisons were made against particular rival agents, not every big name. The 71% figure also leaves out human help, so nobody outside Wajo knows yet how much the people actually add. Early hands-on testing was mixed. A call to an airline about upgrades went well. A call asking a reporter's partner to feed the cats was clumsier, and it was less clear about who was calling. That's normal for a product a few weeks old.

How does an AI agent that hires humans actually work?

An AI agent that hires humans does most of the work in software and passes the stubborn parts to a person. With Fo, the clearest example is the phone call. Most barbershops, plumbers and small clinics have no online booking system an AI can plug into, so someone has to pick up a phone and talk. When that happens, a human assistant on Wajo's staff makes the call. Wajo says that person sees only what's needed for that one task, not your chat history. Poddar summed up the idea in her launch post: other personal AIs pretend AI can do everything, while Fo employs humans for the tasks AI can't handle. That's a refreshingly honest admission in a market built on hype.

Two details matter more than the headline. First, Fo decides on its own when to bring in a human. Going by the founder's description, Wajo covers that cost rather than adding a fee to your task. Second, those handoffs are also lessons. Poddar has said Fo learns from how the humans finish tasks, and that's how it's meant to get smarter over time. Neither point is sinister. Plenty of services mix software and people behind the scenes. But it does mean a stranger may end up holding part of your errand, and how your task gets done feeds back into the product. You'd want to know both before handing over anything sensitive, like a medical appointment or a tricky conversation with a landlord.

The model also raises an obvious question about growth. Paying people to finish tasks works well when a product has a modest user base and a small, motivated team. It gets harder when demand spikes, time zones multiply and the hard tasks pile up on a Friday afternoon. Wajo says Fo is already used in 107 countries but hasn't shared user numbers, so it's too early to say how the human side holds up. Here's my honest take. The hybrid approach is smart for the messy final step of an errand, but it isn't magic. The person helping out is only as good as the brief they're given, and a brief an AI writes for a stranger can lose a lot along the way.

What should you ask before letting an AI agent act or spend for you?

Before any agent acts or spends money for you, ask five things. Who will it contact? Whose name will it use? What will it see? What can it spend? How do you stop it? That's the whole checklist, and it applies to Fo, Muse, Instinct or any other agent that wants permission to act for you. The first question is about who it can reach. Will it only talk to people you name, or can it find and contact strangers, like a handyman it found online? Neither is wrong, but the risks are very different. An agent that only writes to your accountant can embarrass you. An agent that hires and pays strangers can also cost you money, and that's much harder to undo.

Next, identity and access. Does the agent pretend to be you, or introduce itself as your assistant? Fo sends from its own number and email, and that's the right call. An agent that writes from your account blurs who said what and makes mistakes much harder to walk back. Then check what it can see. Does it need your inbox, your passwords, your whole message history? The answer you want is 'only what this task needs'. Finally, look at the money. Single-use virtual cards, like the ones Fo uses, limit the damage if something goes wrong. If an agent wants to keep your real card on file forever, ask why, and ask what the spending limit is before you find out the hard way.

The last question, how to stop it, is the one people forget. You want three controls: a preview before anything goes out, a way to check progress partway through, and a stop button that works instantly. Also ask what happens when the other person says no or doesn't answer. Does the agent back off, or does it keep pinging them until they block you? Agents that nag damage relationships you spent years building. A good rule of thumb: if you wouldn't let a new intern do it unsupervised in their first week, don't let an agent do it without a preview. That isn't distrust of AI. It's the same common sense you'd use with any new hire.

Do you need an agent that hires strangers or one that messages your own contacts?

You only need an agent that hires strangers when the task means finding someone new. For anything involving people you already know, an agent that messages your own contacts and reports back is simpler and safer. Think about your actual week. How many tasks involve finding a brand-new provider and paying them? Now count the ones that look like 'ask Dana when the invoice will be ready' or 'check if the plumber can come Thursday'. For most small business owners and busy professionals, the second kind wins easily. The hard part isn't finding people. It's chasing the ones you already work with, and remembering to do it before the deadline slips past.

That's where the two approaches split. A Fo-style agent is built for open-ended errands: book a table, find a handyman, compare prices, call an airline. A single task might involve a paid human, a virtual card and a business you've never dealt with. An agent that messages your own contacts is narrower on purpose. You tell it who to message and what to find out. It writes to that person, gets the answer and tells you. There are no purchases, no strangers and no human middleman reading your brief. If your work runs on WhatsApp threads with suppliers, clients and contractors, the narrow version covers far more of your real to-do list than the flashy one. Fewer moving parts usually means fewer surprises.

Here's the contrarian bit. In the race to build agents, the most trustworthy feature is often the one that isn't there. Every power an agent has, from spending money to hiring help to calling strangers, is something that can go wrong. Fo handles that with sensible safeguards, and credit where it's due. But if you don't need those powers, you shouldn't carry their risk. Pick the smallest agent that gets the job done. If you're weighing up several options, this plain comparison of WhatsApp AI assistants in 2026 lays them out side by side. Match the tool to the task, not the hype, and look at your choice again in six months, once the market has settled.

How do you get an AI agent to message people for you from WhatsApp?

With Agent Sonic, you tell it in a WhatsApp chat who to message and what to find out. It handles the conversation and reports back. This is the outreach tasks feature. It's in Beta on the Pro plan, so it may be paused from time to time. A typical request looks like 'Ask the plumber if they can come Thursday and tell me what they say' or 'Tomorrow at 10, ask Tom if the contract is signed.' You can share a contact card first, and once a contact is saved you can just use their name. One request can even cover several people, each with their own question, sent right away or at a set time. There's no app to install and no dashboard to learn.

Nothing goes out until you've seen it. Sonic shows you the exact first message for each person. You reply 'yes' to send it, 'no' to drop it, or say what to change in plain words: 'more formal', 'shorter, no emoji', 'call me Sagi'. Those notes stay with the task and shape every message in that conversation, not just the first one, and the approved first message goes out exactly as shown. Sonic writes from its own WhatsApp number and always introduces itself as your assistant. It never pretends to be you or sends from your number. It asks, follows up at most once if the answer is vague, and stops once it has the answer, the person declines, or they ask it to stop.

You get the result as each person's thread ends, whether that's an answer, a decline or no reply, plus a short summary. You can check in any time ('Did Dana answer yet?' or 'What happened with the plumber?'), or send /tasks. /task status shows the latest messages, and /task cancel stops a task. The limits are deliberate. You can run one outreach task at a time and contact up to 3 people a day. First messages go out only between 08:00 and 21:00 in the recipient's time zone, and only in one-to-one chats. Anyone who says 'stop' gets a short apology and is never messaged by Sonic again, whoever is using it. People who don't reply aren't nagged. After 24 hours, Sonic just tells you. This step-by-step outreach tutorial walks through a full task, and you can get access to Sonic by leaving your details. You'll get its number and a quick onboarding.

Frequently asked questions