Do AI SDRs Actually Work? I Was One of 11x's First Customers

Do AI SDRs work? At sending, yes. At selling, no, and the market spent two years and a lot of venture money finding that out the expensive way.

I don't say that from the bleachers. I was one of 11x's earliest customers. I watched the experiment run inside my own pipeline, and I could have told you two years ago how it would end. The tech was genuinely impressive. The premise underneath it was rotten.

What happened to the AI SDR promise?

The AI SDR promise ran into buyers, and buyers won. The AI-SDR crowd spent two years telling you to fire your reps. By August 2026 they had started quietly rehiring them.

11x's CEO reportedly went on CNBC and said AI SDRs don't work, after something like 75% of them churned inside 90 days. Artisan reportedly retired its "Stop Hiring Humans" line, brought on a human BDR, and cut the price from $2,500 to $250. I keep trying to feel gracious about it and keep not getting there.

The pitch never changed the whole time. Replace the people, scale the volume, ten thousand touches before lunch. It held up right until buyers filtered all ten thousand as spam, which is what they were.

Why didn't AI SDRs work?

They were never a new way to sell. They were the old volume game wearing an AI costume: more touches, less human, the same spam buyers had already trained themselves to ignore.

Artisan's Ava 2.0 launch made the point better than I could. The pitch was "replace 90% of your GTM stack," and the proof point it led with was a 7,000-email campaign at 3.6% positive response. That's the number they were proud of. Run the math the other direction and 6,748 people got an email that didn't read them at all. That is the ceiling of spray, printed on the box and sold as a feature.

Fine, automate the 90% that was always going to get automated: the sending and the sequencing. The 10% being waved off is the only part that decides whether a deal closes, which is reading the specific human who has to say yes.

Is the whole autonomous sales category making the same mistake?

Yes, and the rounds keep getting bigger. HockeyStack raised $50M to build "autonomous revenue agents." Attention raised $30M to "run your revenue team, not just record it." Mutiny launched a cold-to-closed GTM agent. Artisan, 11x, and Aligned are in the same race.

Each one came in through a different door. HockeyStack came at it from revenue data, Artisan from outbound volume, 11x from digital workers, Aligned from public signals and CRM data. Different doors, same room, and in that room the agent still fires before it understands who it's talking to.

They all treat a buyer as a persona. Same archetype, same template, same sequence, whoever is actually on the other end, and a lot of them get a little creepy with the "personalization" to boot. Build an autonomous agent on top of persona data and you get an autonomous version of the AI spam everyone is already filtering out. An agent that doesn't know how this specific human makes decisions is spray-and-pray with a better engine and a bigger bill.

When HockeyStack's round landed, my read was simple. The race to the bottom just got $50M faster. Pick a different race.

So should you buy an AI SDR?

Buy one if your problem is that nobody is sending enough. A lot of teams don't have that problem. Founders and the advisors around them ask me this constantly, and that's the whole answer.

None of this means AI loses. AI is great at the volume. What it can't do is read the one human who went quiet in week six and know why. So the industry automated the cheap half of selling, laid off the people who did the expensive half, and acted surprised when the numbers cratered.

The correction is simple. AI works fine. It was just aimed at the wrong half of the job... the send count, when the money is in reading the person who actually decides. If you're comparing vendors on exactly that question, I wrote up where we land against 11x and Artisan.

What to do Monday morning before you sign an AI SDR contract

Run this before any autonomous agent touches a buyer.

1. Pull last quarter's closed-lost and no-decision deals. For each one, mark whether it died because nobody reached the buyer, or because somebody reached them and the deal stalled after the meeting. If the second column is the long one, more sending won't shorten it.
2. If sending really is your bottleneck, buy the volume tool and scope it to sending. Don't fire the people who read deals to pay for it.
3. Ask every vendor one question: how does your agent know that two prospects with the identical title need opposite messages? If the answer is title, seniority, an intent signal, or "our model personalizes," you're buying persona data with an engine bolted on.
4. Put a human read in front of the first message on your highest-value accounts. Who is on the buying team, how each of them decides, and what would make each one go quiet.

That fourth step is the part we built GTM Heroes for. The rep stays the hero of the deal and the product is the sidekick: it reasons about how each buyer, and each person on the buying team, actually decides, then writes that read into the next move. The rep approves it.

I said all of this in public in April. I knew it the day I logged into 11x, two years earlier.