For as long as businesses have had a front door, the cost of walking through it has been paid by the visitor. A person who wanted three quotes made three calls, sat through three hold queues, repeated their story three times. That friction was never a feature, but it did one quiet job well: it rationed the attention of the people who answer.
The inquiry became free
In September 2026 that rationing ended for anyone with a personal assistant. Meta’s Muse launched with payments built in — Stripe Link at launch, Shop Pay across every Shopify store within two weeks, PayPal days later — and the checkout side of the agent economy closed at the top of the market before most businesses noticed it had opened. What the assistant does next is the part that lands on a front desk. It calls. It chats. It fills in the quote form. And because the inquiry is free, one person with one need becomes five parallel conversations: five clinics called about the same appointment, four plumbers quoted for one leak.
Every one of those conversations costs the business the same as a real one. Staff time to answer, a slot held that will not convert, a callback to a person who never rang. The conversion rate per inquiry collapses, attribution breaks, and nobody on the receiving end can say how much of it is happening, because no channel they run carried a label for it.
What a wall costs
The first instinct is the one Amazon acted on. It blocked Muse, and it said why: the agent does not identify itself, and the merchant should be able to decline. Both points are right. But Amazon can afford a wall, because the customer will come back through the front page. A five-person business cannot. The person behind the assistant is a real patient, a real homeowner, a real buyer, and the assistant is how they chose to shop. Refuse the agent and you refuse them, quietly, without ever learning they were there.
Protection sells to the largest merchant on earth. A small business needs the opposite default.
What a small business needs protected is not its perimeter. It is staff time, lead quality, the integrity of the booking calendar, and the right to say no on purpose rather than by accident. None of that is served by a wall. All of it is served by a door that knows who is knocking.
Recognise, then route
A bouncer does three things a wall cannot. It recognises the visitor. It decides which lane they belong in. It keeps a record of who came through. Applied to inbound, that is a label for every contact — a person, an agent acting for one, a person acting for an agent, or a machine — a lane for each kind, and a receipt that says which lane was used and for whom.
Recognition comes first, and it starts with counting. That is why Aethelforge publishes the Index before it sells anything: four channels, one label, a fixed window, a signature. The lane comes next — declared agents recognised and routed into structured intake, undeclared ones challenged rather than refused — and it is built on the same instruments that produce the count.
What we will not build
Three things are being solved by people better placed to solve them, and we intend to consume their answers rather than compete with them.
- Agent identity on the web. Web Bot Auth — signed requests, a published key directory — is verified in production by the largest edge networks and used by the major assistants. We read the header; we do not mint a rival.
- Spend authorisation. Whether an agent is allowed to pay is a question for the card networks and the wallets, who published common rules for it this month.
- The one-vendor business lane. Meta will almost certainly ship a business-side surface for Muse, with disclosures on Meta’s terms. It will be one source among several, and we will plug it in as one.
What is left is the neutral front desk: the place where an assistant from any vendor can ask a small business a question and get a structured, receipted answer, and where the business can see all of them in one count.
The phone is unowned
On the web the identity question has an answer. On the phone it has none. Carrier attestation vouches for the number, not for who is speaking or for whom. No standard exists for “this call is placed by an agent on behalf of this person,” and the first assistant to place calls at scale has done so through human contractors reading a script — so the voice is human and the intent is not. A label that stops at bot-or-human cannot describe that call. Ours names the principal, the intermediary and the intent separately, because the receptionist that answers the line is the one instrument we run end to end.
What we are asking for
If you build an assistant, declare it. Sign your requests, say what you are on the phone, name the person you are acting for. A declared agent gets a lane; an undeclared one gets a question.
If you answer a phone, a chat or a form for a living, count first. Read the Index, then put your own front door on the same instruments. The number that matters is not how many agents are coming. It is how many you can already see.
Aethelforge · an iSimplifyMe project · ChicagoRead the Index