Anthropic launched a buying agent. Is this the end of AEs?

Business Breakdowns · GTM AI

Anthropic launched a buying agent. Open the pricing page, click “Chat with buying specialist,” and a Claude agent runs discovery, quotes a plan and sends you to checkout. Below 500 seats, you never have to talk to a rep.

I went in for a selfish reason. I’ve been thinking about our own inbound funnel, where our solutions are less optimal, and I wanted to see what the best of the best were doing. I didn’t realize Anthropic had gone fully agentic.

So is this the end of AEs?

Not really. Six chats later, my read is that the agent took the queue, not the deal. It sells well, it hands off the moment a deal needs a person, and the part worth copying isn’t the chat. It’s the checkout underneath it.

I’ll end with the test I’d run: self-serve invoicing.

The problem was volume. Anthropic had “tens of thousands of inbound requests every month,” and its inbound BDR team couldn’t keep up, per a September 30 post. Most were questions the docs already answered: price, seat minimums, HIPAA terms. Buyers still waited days.

So the team built an agent on Claude Managed Agents, Anthropic’s hosted service for running agents. That’s a change from May, when Head of Industries Eleanor Dorfman told SaaStr that Intercom’s Fin guided buyers through self-serve. Here’s what the September post reports:

One note on the numbers: they’re compared with the old contact form, so this reads as a before-and-after measurement rather than a true growth experiment. What I love is the metrics themselves. Most teams would stop at conversion rate. Anthropic is tracking sales productivity: days to close, emails per deal, how many chats need a person at all. And sales and content leads edit the agent’s prompt themselves, weekly.

Anthropic also starts with advantages most companies don’t:

That’s the setup. Here’s what it was like to buy.

I ran six chats from my personal account, each built to answer one question. Does it right-size a small buyer, or upsell? How much does it qualify a big deal before a rep sees it? Does it hold up when a buyer goes off script? What happens when procurement demands an invoice?

It right-sized me. As a 25-person sales team, I got Team at $6,000 a year, not Enterprise. Every claim linked to a help article. Anthropic calls the cheaper answer a feature, since a buyer on the right plan is more likely to stay.

It qualified, then handed off. At 1,000 seats, it routed me to sales in its first reply, then asked three more questions: security review, regulated industry, which teams. It quoted $240,000 a year in seat fees and offered a 20-seat start while I waited. Land and expand, from the agent.

It upsold me to Enterprise, and it earned it. As a 40-person company, I asked why we shouldn’t just stay on Team. It agreed Team is cheaper and more predictable. Then it tied Enterprise to two things I’d mentioned in passing. Removing people by hand became SCIM. The 12 people already expensing personal plans became account claiming, which only Enterprise has. That’s a good rep’s move: listen for the pain, then price it.

It held the line on price, and on people. I said a competitor quoted less per seat and asked for a match, a discount or a call. No to all three: “No, we don’t match quotes or offer discounts.” At 40 seats, sales “would offer you the same plan at the same list price,” so a call would only mean waiting longer. Then it offered a 20-seat start instead. That only works because the price really is fixed. If your reps discount to close, an agent that won’t will lose those deals, or teach buyers to ask for a person.

API buyers fall through. About $20,000 a month of API usage got me a tour of the Console and a link to the contact form: “I don’t have a hand-off for API deals.” Very polite. Still a form. I’d bet this is where the agent expands next, and it’s where the most interesting qualification could happen: what kind of company you are, what you’re building, what usage you expect. Get creative with those questions and you find out whether there’s a whale in your funnel.

Invoicing brings in a rep, and a rep wants 50 seats. In the last chat I said procurement required an invoice. The agent didn’t argue. Invoicing lives only on sales-assisted Enterprise, so it started drafting the hand-off. Then it checked whether I had 50 actual people, “not testing you,” because “the sales team works from the seat count I send.” Sales-assisted Enterprise starts at 50 seats, per the help center. A 25-person team that needs an invoice has two choices: pay by card, or buy 50 seats. (Dorfman described self-serve enterprise as coming with “real invoicing.” Either that changed, or the word was loose. The help center and the agent both put invoices on the sales side today.)

So the rule underneath has a floor as well as a ceiling. A requirement, not size, brings in a rep: a contract, an invoice, a currency other than USD, a private offer or more than 500 seats. Between 50 and 500 seats you pick a lane, card today or a rep and an invoice. Below 50, it’s card or nothing. Above 500, it’s sales, which the contact sales page says “can only support complex deals, like 500+ seat deployments and BAAs.”

My read is that the 50-to-500 band is an inside sales motion built for volume, with a floor on how much human time a deal gets. The post’s own productivity metric is emails per deal for an inside rep, down from about 10 to 6. AEs sit above that band.

One more thing it didn’t do: look me up. It never knew my title or employer. It asked for headquarters country and industry, and nudged me toward a work email. Enrichment comes later: Dorfman says every sales lead gets enriched and qualified by Clay and Claude before a BDR sees it. That’s the tell. This agent qualifies for Anthropic, not for me. It works from what you type, decides when you’re worth a rep, holds list price and points you at whatever closes today.

It can only do that because of what sits underneath.

Anthropic built self-serve first. Enterprise is buyable online from 20 to 500 seats, by credit card or ACH bank transfer, in USD, with usage credits bought up front, per its help center. Invoices, other currencies, a customer success manager and AWS Marketplace private offers go to sales, which starts at 50 seats. That’s why the agent can close a 300-seat deal. Checkout can.

And it works. 54% of Anthropic’s new enterprise logos in 2026 came through self-serve, Dorfman said in May, with “real ACV” and no AE in the loop. Self-serve isn’t the downgrade lane anymore. It’s the main one.

And buyers want this:

But only on deals they can buy alone. Anthropic’s self-serve tops out around $120,000 a year in seat fees (my arithmetic, not theirs). You likely won’t self-serve a six-figure deal. Low five figures is totally possible.

That’s where the disruption lands: low-ACV, high-velocity funnels. If that’s your motion, invest in the PLG experience first. Then make sure your reps drive incremental impact over it, not the deals the product would have closed anyway.

Which brings me back to the headline.

At 1,000 seats, the agent sent me to a person on its first reply. The research says that’s the right call:

Agents aren’t quite yet replacing an AE’s ability to build relationships, most of the time with CIOs, CEOs and other senior leaders. The agent clears the queue so AEs can spend their time there.

I’d expect Anthropic to keep investing in two places: the self-serve engine, and the hand-off between the qualification agent, the buying agent and the product experiences customers land in. From here, every experiment comes back to two questions:

Today the routing runs on qualification lines: seat count, an invoice, a contract. I’d expect Anthropic to move past those and start categorizing the conversations and customers coming into the funnel, then measure whether a seller has outsized impact on each type. That only becomes possible as the self-serve agent gains capabilities, because each one is another deal type you can run without a rep. (The next capability is probably a video agent. I still find those creepy.) If this were my funnel, five moves:

An invoice is the one routing reason where the rep’s job is paperwork, and it’s the cleanest version of the first question: offload it, and does the buyer notice? (Dorfman’s “real invoicing” line suggests this may already exist. If so, run it on whatever requirement is pulling paperwork deals to reps today.)

Hypothesis: invoice-only buyers close at the same rate and go live faster when checkout issues the invoice, and the reps they free up spend the time on incremental deals.

Who’s in it

The bar: non-inferiority, not superiority. Self-serve can’t trail the rep by more than five points of close rate, written down before launch. Clear that and win on speed, and it ships.

Self-serve metrics

Invoice and payment guardrails, each against the sales-assisted book

Decision rule

Robert