The AI SDR hype is over. Here is what actually survived

The first wave of AI SDRs sold a simple idea: replace your sales development team with software that never sleeps. That idea mostly failed. What survived is narrower and more useful: agents doing the work that scales, and people making the calls that carry risk.

What the AI SDR wave promised, and what broke

The pitch had three parts: replace the team, send without limit, and run on its own once set up. Each part broke on contact with real buyers.

Replacement. In an informal SaaStr reader survey in late 2024, 83% said they had gotten nothing from AI SDRs yet, and only 3% reported real revenue.

The best-known story is 11x, an AI SDR startup backed by Andreessen Horowitz and Benchmark. In March 2025, TechCrunch reported that 11x had shown logos of companies that said they were not customers, and that most early customers used break clauses, typically at three months, to leave. A former employee said some customers had hoped to replace “an entire outbound sales team.” 11x disputed parts of the reporting, said its retention was 79%, and said performance “ultimately relies on the quality of user input.”

Remember that last line.

Volume. Cheap writing made cheap sending, and buyers noticed. In Gartner’s survey of 632 B2B buyers, 73% said they actively avoid suppliers who send irrelevant outreach. Belkins’ study of 7.5 million cold emails shows the crowding: reply rates fell from 0.50% in the first half of 2025 to 0.40% in the second, a 20% drop within one year, which the study blames on saturation, stricter filtering and finite attention.

Set and forget. SaaStr reports 15 to 20 hours a week of human oversight for its own AI SDRs, and says it bluntly: “This is not set-and-forget technology.”

Notice what the three failures share: each treated outbound as a production problem (more emails, fewer people) when it is mostly a judgment problem (who to write to, why now, and what to say when they answer). Software made production nearly free. It did not make judgment free, and that is where everything that survived lives.

Survivor one: proven playbooks and real data

SaaStr’s first lesson is that AI SDRs “scale what’s already working” and cannot fix what is broken. If outbound did not work when people did it, an agent will not rescue it. It will do the wrong thing faster, in more inboxes.

Most readers skip one detail: SaaStr’s outbound agents mostly wrote to people already in its own database (“About 90% of contacts are ours”). Strong numbers on warm contacts say little about cold outreach, so before you benchmark against any case study, ask how cold the list really was.

That is what “the quality of user input” means in practice. An agent is only as good as what it is told about the market, the offer and the buyer. That input is a playbook, and someone has to have proven it first.

Survivor two: a reason to write now

The emails that still work carry a reason that belongs to this person, this month: a new site, a new leader, a new market. A signal answers the question every recipient asks before replying: why me, and why now? The Belkins study ends on the same note: “relevant message, right person, right moment.”

Compare two openings to the same operations leader, both invented for this example: “I help logistics companies cut costs,” and “You opened a second warehouse in Ohio in May, and a new site usually means a new set of fuel cards.” Only the second gives the reader something to answer, and only the second could not have been sent to everyone else on the list.

Survivor three: people at the decision points

The setups that hold up put people exactly where a mistake would be expensive. SaaStr advises starting agents in “draft mode,” where the agent suggests and a person approves, and six months in it still kept high-value prospects there.

Buyers want the same thing. In Gartner’s 2026 survey of 645 B2B buyers, 69% said they prefer to validate AI-generated insights with sales reps. An agent can open the conversation, but a person still has to be reachable when it matters.

Survivor four: inbound first

The biggest win in SaaStr’s account was not cold outbound at all. It was an inbound agent that answers website visitors and books them in seconds to minutes, where the old handoff took one to three days. In October, according to SaaStr, 70% of its closed revenue came through that agent.

The honest lesson for anyone shopping for outbound, including from us: if inbound leads wait a day for a reply, fix that first. People who already raised their hand are the cheapest meetings you will ever book.

What to do next: five questions for any AI SDR

Before you sign with any AI SDR vendor, ask these and get the answers in writing:

1. Who approves the plan and the first emails before anything is sent? If nobody does, your brand is the test environment.
2. What happens, mechanically, when someone replies “stop”? Listen for a rule in the system, not an instruction to the model.
3. Whose domain sends the emails? If it is yours, a bad month lands on the address your customers and invoices come from.
4. What do you pay if no meetings come? That tells you who carries the risk, whatever the brochure says.
5. Does the agent say it is an AI, and who it works for? If not, ask what happens to your name when a prospect finds out.

A vendor who answers all five clearly has learned from the first wave. Our answers are in how Dylan is built and why he charges per held meeting.