Guide 50 · Basics

What Is Wajo? The AI Agent That Phones Businesses for You

Wajo's Fo is an 'action agent' from an ex-DeepMind engineering lead: it books, orders, pays, and actually calls businesses on your behalf — and hands the job to a human assistant when it gets stuck.

By Abhit Tomar · Updated October 20, 2026 · 20 min read

Editorial illustration of a small robot assistant holding a vintage telephone handset, with checkmarks, a calendar, and a credit card floating around it
Illustration: museaicodes

Short answer

Wajo is a startup founded by Shivani Poddar, a former engineering lead at Google DeepMind. Its flagship product, Fo, launched September 29, 2026 as an 'action agent': instead of just answering questions, Fo makes phone calls, sends emails, books appointments, places orders, and processes payments on your behalf — and when it can't finish a task, a human executive assistant completes it rather than letting the AI guess. Wajo's own fieldnotes claim 71% task completion vs 42% for OpenClaw, but those figures are self-reported, not independent testing. This is an independent explainer — museaicodes.com is a Meta Muse guide site covering Wajo because anyone choosing an AI agent today is choosing between all of them.

Watch: Fo's use cases — calling, email, booking, payments, Fo MCP, invites

What is Wajo (Fo), in plain terms?

Wajo is a startup founded by Shivani Poddar, who spent years as an engineering lead at Google DeepMind. Its flagship product is Fo, an AI 'action agent' that launched on September 29, 2026. The pitch is a reaction to a specific frustration: AI can already answer almost any question, yet the calling, comparing, booking, and coordinating still lands on you. Fo exists to take that half of the work.

In practice, Fo is the agent you hand an errand to instead of a question. 'Dispute this bill,' 'book a table for Friday,' 'find me a plumber for the weekend' — Fo is built to carry those through to completion: calling the business, sending the email, placing the order, paying with a single-use card. It connects to your Google Workspace so it builds context about you over time, and it asks for your approval on important decisions rather than acting silently.

The idea that separates Wajo from every other personal agent on the market right now is what happens when Fo fails. Most agents guess, hedge, or quietly stall. Fo escalates to a human executive assistant who finishes the task. Wajo's bet — in the founder's own framing — is that a mixed team of software and people beats software alone, at least for now.

How Fo actually works: the human-fallback loop

Every errand Fo takes runs through the same loop. Understanding the loop matters more than the feature list, because the loop is the product.

  1. You hand Fo an errand in plain language — the kind of task you'd otherwise do yourself across three apps and a phone call.
  2. Fo works it the way an assistant would: checking your connected accounts for context, sending emails, filling forms, booking services, and placing phone calls to businesses where there's no API to talk to.
  3. Payments go through single-use cards issued per purchase, so no merchant ever holds your real card number.
  4. If Fo hits a wall — a phone tree with no good options, a request outside its authority, a judgment call it shouldn't make — it stops and hands the task to a human executive assistant instead of improvising.
  5. The human sees only the information necessary to complete that task, not your full chat history, per Wajo's published design. You get the finished errand back.

What Fo can do

  • Place outbound phone calls to businesses — the capability no other leading personal agent currently offers (per October 2026 hands-on testing of Muse, Instinct, Wajo, and OpenAI's Dots).
  • Send emails and coordinate with vendors on your behalf.
  • Book appointments and services, including travel changes.
  • Place orders and process payments via single-use cards per purchase.
  • Connect to Google Workspace to build context about you over time — preferences, ongoing threads, dropped follow-ups.
  • Catch the small expensive mistakes: early users cited by the company report Fo catching a forgotten free trial (saving $80) and finding a 25% food-order voucher.

Getting started

Wajo's official site is wajo.ai — that is where sign-up and the current availability details live, and it's the only source you should trust for the live state of the product. Wajo has not published a detailed setup walkthrough, so treat anything beyond the official site as unverified.

What launch coverage does confirm about onboarding: you connect your Google Workspace so Fo can build context about you over time, and you grant it the communication channels it needs — email, and phone capability for the outbound calls. Fo asks for your approval on important decisions, so expect an approval step before anything consequential.

The sensible way to start with any agent this new: one low-risk errand first. A restaurant booking or a bill question, not your primary inbox or your main payment method. Widen access as it earns trust — the same rule we recommend for every agent in this category.

Signups run through invite links — the app's Settings page has an Invites tab for sharing your own invite link. Don't have one from a friend? Use ours: sign up for Wajo with our invite link — it helps us keep the guides free.

Pricing

Fo is free today, per October 2026 coverage. That's the whole pricing story as published — Wajo has not announced paid tiers, usage limits, or what 'free today' converts into later.

Read that honestly: free at launch for a week-old startup is a growth tactic, not a pricing page. If you're evaluating Wajo against agents with published pricing, price it as 'unknown' rather than 'free' — the number that matters is the one you'll pay in six months, and nobody outside Wajo knows it yet.

First-hand details from an actual account (October 2026): the plan is called 'Fo mini' with a daily limit of 1,000 tasks per day. The settings page has a 'Have a code to upgrade?' field — invite/upgrade codes unlock higher tiers. Support is at support@wajo.ai, and you can permanently delete your account along with everything your agents have stored.

Fo's code execution limits, in its own words

Ask Fo about its own workspace and it gives concrete numbers: about 4 GB of memory total for running code (roughly 3 GB usable at once), 10 GB of durable disk storage that persists between sessions plus 10 GB of scratch space wiped on restarts, a 20 MB cap per output file, and a 15-minute cap per code run.

One honest caveat, from Fo itself: when asked whether those numbers are per-user, it declined to say, noting it doesn't disclose how the platform provisions resources per account. Treat them as the workspace Fo describes — useful for sizing your tasks — not a published SLA. And as Wajo's own footer puts it: agents can make mistakes, so double-check.

Safety, privacy, and honest limits

Start with what Wajo claims about itself, labeled as such. In the company's own published fieldnotes, Fo completed 71% of 104 real-world tasks without human fallback (vs 42% for OpenClaw, the open-source assistant Wajo used as its benchmark), and passed 94% of its safety checks — tests for contacting people without authorization, leaking private information, or making unauthorized decisions (vs 74% for OpenClaw). These are Wajo's figures from Wajo's tests. They are the company's pitch, not an independent review, and you should read them that way.

The privacy design has one genuinely thoughtful piece: when a task escalates to a human assistant, that person receives only the information necessary to accomplish the task — not your full chat history. That's better than the obvious alternative. It is still a stranger reading part of your errand, and Wajo is a weeks-old company whose data-handling practices have not been tested by time, regulators, or security researchers.

The honest limits, stated plainly: availability by region is not clearly published. There is no public information on which AI model powers Fo. The human-fallback network's capacity — what happens when thousands of users escalate at once — is unproven. And a 71% completion rate, even taken at face value, means nearly a third of tasks still need the human backup or your own intervention.

None of that makes Wajo a bad product. It makes it a young one. Judge it on finished errands, not on launch-week numbers — and keep consequential tasks supervised until it has a track record.

Wajo vs the alternatives

Muse (Meta) is the agent you can install and evaluate today — public pricing, documented safety architecture, and warranty-backed payments. Our planned Muse vs Wajo head-to-head will score the two errand-for-errand once both can be tested side by side; until then, start with What Is Muse AI? for the full picture of Meta's agent.

Instinct is the text-message-native rival that raised $1B at a $10B valuation — the most hands-off of the group, but still invite-only. Our Muse vs Instinct comparison covers how a finished product stacks up against a promising waitlist.

OpenAI's Dots are the always-on agents inside ChatGPT for Pro and Business users — the strongest pick if you already live in ChatGPT, Slack, or Teams. See our launch coverage on /news and the Dots vs Muse deep-dive.

Real-world use cases: what Fo actually does

From a live account (October 2026), Fo's home screen greets you by name and offers two one-tap starters — "Make a call for me" and "Book an appointment / service" — under a simple "Ask Fo anything…" input. Fo also has its own contact points: fo@wajo.ai and the phone number +1 (628) 309-6361, so it can reach businesses — and be reached — like a real assistant.

The use cases that fall out of that design: billing disputes and customer-service calls where you'd otherwise wait on hold; appointment and service booking (clinics, salons, repair visits) including travel changes; and vendor coordination where Fo emails, calls, and follows up until the errand closes. The "human-fallback loop" means the task doesn't die when Fo hits a wall — a human executive assistant picks it up.

  • Phone errands: Fo places outbound calls to businesses — the signature capability. Think disputing a charge, rescheduling a flight, or getting a plumber out for the weekend.
  • Booking and coordination: appointments, services, and travel changes, with email follow-ups to vendors handled on your behalf.
  • Agent spending: the Payments settings let you link a card so Fo can make purchases for you. Card details are held by Stripe — Wajo says it never sees or stores the full number.
  • Spending guardrails: "Ask before purchases over $X" (amounts in USD), a "Pause agent spending" toggle that stops all purchases, and a Balance line showing any outstanding amount.

Fo inside ChatGPT and Claude: the Fo MCP server

The most interesting new surface in the app is Settings → Fo MCP: "Fo in your other apps." It connects Wajo's assistant to your Wajo account so you can "ask in ChatGPT or Claude, and Fo gets it done in the real world." The screen offers tabs for ChatGPT, Claude, and Claude Code.

Setup is a server URL plus OAuth. Wajo gives you the MCP endpoint — https://mcp.api.wajo.ai/fo — and a three-step flow: open ChatGPT's plugins screen and choose Add → Add custom MCP server; add Fo as a plugin (name it Fo, paste the server URL, select OAuth); then sign in with Wajo — "Choose Create as a plugin and complete authorization with Wajo. Review the tools, then pick Fo in your chat."

In practical terms, this makes Fo a portable action layer: you keep your preferred chat interface (ChatGPT, Claude, or Claude Code) and Fo shows up as the plugin that can phone businesses, book, and pay. It's the one piece of the Wajo story that also matters to developers — Claude Code in particular turns Fo into real-world tooling inside a coding workflow.

Invites

Wajo's growth loop lives under Settings → Invites, where each user gets a personal invite link to share with friends. New signups arrive through these links.

Don't have an invite from a friend? Use ours: sign up for Wajo with our invite link — the referral helps us keep the guides free.

Watch, listen, or read: Fo's use cases

Prefer video, audio, or a printout? We made three companions for this guide. The video above walks through Fo's use cases — calling, email, booking, payments, the Fo MCP connection, and invites — in about 70 seconds.

The audio below is the same walkthrough as a listen-alone narration (71 seconds).

And the two-page cheat sheet (PDF) compresses everything onto paper: the four use-case groups, the Fo MCP setup steps with the server URL, the invite link, and Fo's contact details (fo@wajo.ai, +1 (628) 309-6361).

Listen: Wajo Fo use cases in 71 seconds

Sources & further reading

Keep exploring

Comparison of Muse AI, ChatGPT, Claude, and Meta AI
AspectDetail
Maker / founderWajo — founded by Shivani Poddar, former engineering lead at Google DeepMind
LaunchedSeptember 29, 2026
ProductFo — an 'action agent' that completes errands (calls, emails, bookings, orders, payments) rather than just answering questions
AccessFree today; signups are invite-based ([our invite link](https://wajo.ai/join?code=2VJ5NW5G&utm_source=referral&utm_medium=invite_link)); regions and waitlist mechanics not clearly published — check wajo.ai for the live state
PriceFree today (October 2026 coverage); no paid tiers or future pricing published
Key capabilitiesOutbound phone calls to businesses; emails; appointment/service booking; vendor coordination; travel changes; orders; single-use-card payments; Google Workspace context
Human fallbackWhen Fo can't finish, a human executive assistant completes the task — the human sees only task-necessary info, not full chat history
Safety story94% on Wajo's own safety checks (self-reported, not independent); approval asked on important decisions; data practices of a weeks-old company, untested at scale
Best forErrands that end in a phone call — billing disputes, bookings with no online option, contractors — where other agents leave you to finish the job yourself