Executive summary: On November 12, 2026, Meta's consumer AI agent, Muse, became able to place calls to US businesses on a user's behalf — for paid subscribers. Until then (a September 2026 beta), only a select group of early adopters could ask Muse for the feature. Now any paying Muse subscriber in the US can request an outbound call. The task it owns is the mundane one: the cable company, the DMV, the insurer, the restaurant that does not take online reservations. Meta's chief AI officer, Alexandr Wang, put it plainly: "let muse take care of all those calls you don't want to make." For a small business, the transferable capability is not Meta's specific product. It is the pattern: an agent with a voice, pointed at the data you already have, with a human gate on the money steps, running on a cost guardrail and a cadence. The playbook below is four steps: name the call you are always dodging, give the agent the context, gate the money steps, and schedule the phone desk. EP55 of The AI Guy Podcast walks the same four moves in audio.
Every owner is a human router. You copy data between the phone and the tools. You chase the invoice that is six weeks old. You call the supplier to confirm the order. You sit through a four-minute phone tree just to reschedule one appointment. Each call is a small tax on your afternoon, and the tax never ends, because the phone desk never closes. When you are a team of two, you do it yourself. When you are one person over, there is nobody else to give it to. So it stays on you. The phone is where your best hours go, and where your customers quietly leave: the missed after-hours call, the callback nobody returns, the hold that runs past the patience. The obstacle is not that calling is hard. It is that calling is always on, and you are the only one who answers it.
Tools & pricing: Meta Muse (consumer AI agent; calling is a paid feature, plans start at $20/mo for 500M tokens/week, up to $100/mo for 3B tokens — verify current pricing and US availability). Competitors with similar capability: the startup Instinct, and Google's Gemini. The capability — an agent that dials, navigates a tree, and holds — is the part you should replicate at SMB scale, not necessarily the specific product.
This is the whole move in one sentence. Not all the calls. The single routine outbound call that eats the most owner time. The invoice chase. The price check. The reschedule. The supplier confirm. Write it down. One call. That is the target your agent gets to own. The reason this is step one and not step four is that a phone agent without a named target is a phone that calls things at random. Give it the specific call it should keep making, and it has something to do. The call you are most likely to keep postponing is, by definition, the one most likely to quietly hurt you: the invoice you keep meaning to chase, the appointment you keep meaning to confirm.
Tools & pricing: A notebook, a whiteboard, or the call log in your CRM. Free. The cost is the five minutes it takes to name the call honestly.
A voice that dials is half the job. The other half is context. A phone agent that does not know what it is calling about is worse than none, because it will confidently say the wrong thing. Point the agent at the one system of record behind the call: your CRM, your invoice list, your booking calendar. The mechanism is the same as any agent: it reads what you already have, and answers in the context of that data. When it calls to confirm order number 429, it already knows what 429 was, who it was for, and when it was due. No new system required to start. Do not wire up twelve tools on day one. One system of record, one number, one call. If the agent can tell you, in plain terms, why the invoice for that account is still open, you have more signal than you have had in a while.
Tools & pricing: Your existing CRM or billing tool, plus a voice-capable agent (Meta Muse's paid tier, or a general assistant with telephony access). No new system required to start. Note: a call inherits whatever context the agent's connectors can already read, so decide deliberately what it can see before it dials.
This is the part that keeps you safe, and the part that separates a tool you trust from a liability you clean up. The agent dials, it listens, it gathers. But anything that spends money, commits the customer, or changes a contract stops and comes back to a human. You are not building a robot that negotiates your business away. You are building a robot that does the ninety percent of the call that is boring, and hands you the ten percent that matters. The gate is not a limit on the agent. It is what makes you trust it enough to let it dial in the first place. This is also where the privacy question lands: the calls the agent makes inherit its context, and a Meta trial that routed some calls through human contractors was suspended over privacy-disclosure concerns. The principle is the same either way: you control what the agent may spend, commit, or share, and a human signs the money step.
Tools & pricing: A rule, not software. Cap what the agent may spend or commit, and route the money step to a person. The rule is what matters; the tooling is just where you write it.
An agent you have to remember to use is a phone nobody answers. The move is to make the phone desk automatic. Scheduled. Every morning, the agent runs the list: chase the invoices, confirm the orders, book the appointments that were missed. You review one page. What got done, what needs you, what it could not reach. Five minutes, before you open the rest of your day. The cadence is the point. On-demand calling is a task; a scheduled phone desk is a habit, and the habit is what changes the decision you make. And because the agent runs in the background, the marginal cost of running it daily is close to zero once the guardrail is set — a monthly token budget, with a hard stop, and the hard calls routed to the stronger model.
Tools & pricing: A calendar or a scheduler + your agent. No new software. The cost is five minutes a day, plus the monthly token budget.
Three things worth stating plainly. First, calling is a paid, US-only feature for now; it is not on the free tier, and Meta has not published a success rate, an average call length, or a list of businesses that accept agent calls. Second, reception on the other end can be mixed: many businesses and receptionists still hang up the moment they realize it is not a human on the line. Better voice mimicry helps, but it does not guarantee the call connects. Third, an agent is only as good as the data and the gate behind it. If the context is wrong, the agent will confidently make the wrong call; if the money step has no human gate, the agent will spend and commit without you. The agent does not remove the need for judgment. It removes the need to be the last person to make the call.
The direction is clear. The agent is moving from the text box to the phone, and the next version is the agent that makes the call, takes the callback, and works the queue while you sleep. The companies that win will not be the ones with the smartest agent. They will be the ones who stopped being the last person to call. And the realistic ceiling is honest, not dramatic: it returns hours, not transformation. Twenty dollars a month is a small number. The hours are the real number. If the agent takes thirty of your calls a month and each was worth five minutes of your time, that is twenty-five hours back. That is not a productivity stat. That is your week, partly, returned. Baseline it this week: write down the one call you are always dodging, point the agent you already have at the data behind it, set the spend guardrail and the human gate on the money steps, and schedule the phone desk for Monday morning. The fastest business wins the week. The one that stops making the call itself wins the year.
Your next move is one call. Name the routine outbound call you keep postponing this month. Point your existing agent at the data behind it, set the usage cap and the human gate on the money steps, and schedule the phone desk for Monday morning. To get the full SMB agentic-AI strategy, including how the phone desk stacks with the in-suite agents and where the time-reclaim actually pays for itself, start here:
Get the SMB AI adoption strategy