SaaStr实战：21个AI代理关闭数百万营收，但尚无合格AI AE

We run 21 AI agents in production at SaaStr. They have closed millions in revenue. They book meetings on Saturday nights, resurrect leads nobody had touched in six months, run the invoice and the collections follow-up, and write to Salesforce all day without asking anyone for permission.

It’s so great and so disruptive, and now our human GTM team of 1.5 or so does as much or more than 6+ used to. And yet, in many ways, we have barely gotten anywhere yet.

Agents can already replace most of the SDR function. They can replace a lot of first-line support. They can replace a meaningful slice of CSM work. What they cannot do yet, in any product I have deployed or seen demoed on our own stage, is close a real deal.

That is going to change outside of field sales and complex enterprise. But right now, the explosion of self-serve and agent-serve for AI products has hidden a gap most founders have not priced in: there is no good AI Account Executive yet.

GTM Agents Are Great. But to Get to The Next Level, They Have to Close, Too:

• Every dollar our agents have closed came from a motion where the buyer was already leaning in. Inbound, win-back, renewal, self-serve. None of it came from an agent talking a hesitant buyer into a decision.
• Agents take the easy half of any function first. Pylon’s data on support says it plainly: deflected tickets are the easy tickets. Sales works the same way, and what is left for the AE is the hard residue.
• PayPal ran Agentforce against ~8,000 leads a month no human was going to call and lifted meeting conversions 50%. That is agents fixing coverage, not agents closing.
• The SDR compression is real and measured. Emergence’s survey of 560+ B2B companies found 36% cut SDR headcount, only 14% cut sales engineers, and 28% grew AEs.
• Self-serve is not an AI AE. Anthropic closing 54% of new enterprise logos self-serve is a great result, and it works by removing the need for a closer rather than automating one.
• The first place agents will truly close is anything that can close over text. That is a bigger share of mid-market than most founders admit.

#1. What Our Agents Actually Closed, and How

Agent revenue is a category that does a lot of work in most pitches, so here is exactly where ours came from.

Our inbound agent (Qualified) has closed over $1M in sponsorship revenue. It got there by handling 442,000 chats and converting them into 614 booked meetings. The agent qualified, routed, and scheduled. It did not negotiate a rate card.

Our win-back campaigns through Agentforce hit 72% open rates on about 1,000 ghosted sponsor leads, with 10%+ response rates, and produced closed deals from contacts written off half a year earlier. That is an agent doing the single hardest thing about follow-up, which is doing it at all, forever, without getting bored or discouraged.

Our AI SDR layer sends around 3,200 emails a month. A good human SDR at our size sent 75 to 285. That is a different unit of work, not a better version of the old one.

And 10K, our AI VP of Marketing, now runs the back half of the deal end to end. When a PandaDoc signature comes in, it flips the opportunity to Closed Won in Salesforce, appends missing contacts, creates and sends the bill.com invoice, and runs collections reminders with 7-day escalation. It proposed running commission calculations itself, and now does that too.

The agents open the deal and they process the deal. The moment in the middle where somebody decides to spend money is still handled by a human.

#2. Agents Always Take the Easy Half First

The sharpest version of this at SaaStr AI 2026 came out of a support session. Pylon’s Marty Kausas and Advith Chelikani showed a roughly 5,000-person company with about 1,000 people in support deflecting around 50% of tickets with zero headcount change. The reason headcount did not move: deflected tickets are the easy ones. Removing them does not remove a person, it concentrates the remaining humans on harder work.

Klaviyo’s Andrew Bialecki described the same pattern from the build side. Their agents train up to 50-70% resolution and then hand off. The handoff point is where the value of the human starts.

The sales version of this showed up in the Selling SMB With Agents session, where Amelia Lerutte sat down with Adam Alfano, President at Salesforce, and Eitan Saban, Head of Sales for North America Mid Market at PayPal. PayPal put Agentforce on roughly 8,000 leads a month that no human was ever going to call, and meeting conversions went up 50% inside 14 weeks. That is a large agent-driven revenue result, and what it is is coverage of pipeline that was being abandoned. The agent did not take deals away from AEs. It worked the deals the AEs were never going to get to.

Sales has exactly this shape, and it is why “agents closed X” numbers can mislead. Agents are absorbing the top of the funnel, the mechanical middle, and the paperwork at the end. The part they leave behind is the part that was always hard: a buyer who is not sure, a champion who went quiet, a procurement team that wants terms nobody has offered before, a competitive bake-off where the answer depends on reading a room.

That is the AE job. It is the residue of everything that could not be automated.

#3. Why Closing Is Structurally Harder Than Qualifying

Qualification is a classification problem. Follow-up is a scheduling problem. Quote-to-cash is a workflow problem. Agents are good at all three.

Closing is a judgment problem under ambiguity, with authority attached. Four things make it different:

• Concession authority. Closing requires the ability to give something away and be accountable for it. Nue’s James McArthur showed why teams are nervous about handing that over: their quoting AI is deterministic on purpose, with discount guardrails enforced at the line-item level so a 76% discount request gets capped at 55%. The system is built so the AI cannot decide the price. That is the right design today, and it also means the AI is not closing.
• Reading silence. Half of closing is interpreting what did not happen. No reply from the champion, a new name added to the thread, a procurement question that signals the deal is real. Agents currently treat silence as a trigger for a follow-up sequence, not as information.
• Multithreading with memory. A real deal is six people, three months, and a dozen artifacts. Agents get magical when they hold context no human could hold, which is exactly the argument for why this eventually flips. They are not there yet on the political layer.
• Being wrong loudly. We have lived this. An agent guardrail we did not ask for once skipped a signed contract because the deal title did not match an expected string, and quote-to-cash broke silently. In sales, silent failure is the worst possible property. A human AE who is losing a deal usually tells you.

#4. Self-Serve Is Not an AI AE

The thing hiding the gap is that AI companies are growing fast without AEs, so it looks like the AE problem has been solved. It has been sidestepped.

Anthropic’s Eleanor Dorfman rebuilt the sales org in January 2026 and four months later 54% of new enterprise logos were closing self-serve. Gamma’s Grant Lee got to $100M ARR with almost no sales team, and his own read is that they reacted rather than planned, and he would advise against that. Vercel’s Jeanne DeWitt Grosser described a lead qualification function that went from about 10 people to roughly one.

None of these is an AI closing a deal. They are companies removing the need for a closer on the deals that never needed one, and then still building conventional sales teams for the ones that do. Anthropic’s careers page in late May 2026 had 72 open sales roles against 67 AI research and engineering roles. Sales was around 20% of all openings.

Every AI-native company that gets big enough ends up with a sales team. AI changed the timing, not the destination.