AI SDRs: do they actually work in 2026?

Sunny Ray · Sep 24, 2026 · 6 min read

AI SDRs work for some things and fail at the thing they're sold for. As tools for research, list building, drafting and handling replies, they save real time. As a full replacement for a human doing outreach, the results in 2026 are thin and often bad: buyers now get so many AI-written emails that the generated "personal" touch reads as noise. The teams getting results keep a human in the loop.

A GTM consultant said it best when he was a guest on my show yesterday. The early adopters were "absolutely smashing it" in 2023 and 2024. Now everyone's getting hundreds of AI emails a day, and the biggest mistake companies make is taking the human out of the loop entirely. Here's the longer version.

What an AI SDR is

An AI SDR is software that does the job of a sales development rep: find prospects, research them, write outreach, send it, follow up, handle replies and book meetings. Some do one or two of those. The ambitious ones promise all of it, with no human required.

That last promise is the part to be careful with.

Why the first wave worked

When AI SDRs first showed up, they had an unfair advantage: nobody else was using them. A well-researched, relevant email at scale was rare. Buyers got a handful of cold emails a week, and the AI-written ones were often better than the human-written ones.

My guest put it plainly on air. In 2023 and 2024, "if you were early to the table with like the quote unquote AI SDR," you were absolutely smashing it.

Why it stopped working

Then everyone bought one.

In his words, "we're all getting, you know, 300 emails a day and the context is like, hey Sunny, saw you host a live show." The context is forced. It's technically personal and obviously automated. He made a point I loved: you and I would never reach out to each other and comment on something the other person posted two years ago. Nobody talks like that. So when an email does, you know a machine wrote it.

His bigger point was about the human. Context matters, but so does taste and judgment applied to that context. The biggest mistake he sees is removing the human from the loop entirely. And for a company still in founder-led sales, the founder can talk about the business better than anyone or anything right now.

When I prepped for that episode, my own note to him said most go-to-market AI ends up as "a faster way to send worse messages." The episode only made me more sure of it.

What independent data exists (not much)

Here's the honest state of the evidence: there isn't much. Instantly, which publishes a lot of cold email benchmarks, states there is no AI sales agent specific benchmark yet.

What they did publish is a survey of 250 agencies. 41% have at least one AI agent in production. And agencies that skipped evaluating their workflows were below break-even, at 0.7x return on what they spent. In other words, the tool isn't the result. Testing and supervising it is.

On the other side, the horror stories are easy to find. Prospeo collected Reddit threads where users describe AI SDR output as "more volume, less signal," and one user of a well-known AI SDR product called it a "literal disaster." That's anecdote, not data. But it lines up with what I hear on calls.

Where AI helps: research, replies, scoring

I'm not anti-AI. We run on it. The question is where.

• Research. AI reads a prospect's company, role and recent news in seconds. That's pure upside.
• Drafting replies. Every call I take is recorded, and I can ask my AI to draft a hundred follow-ups to whoever needs one. It's better at that than I am, and I still read them before they go.
• Scoring. AI can rank who's most likely to be a fit, so the human time goes to the best leads.
• A small daily batch. My guest's own workflow is a good model: the system puts together the context for each prospect, it shows up every morning, and a person runs through five or ten a day. For bigger-ticket deals especially, that review is where the value is.

That's the pattern: AI does the work around the human, and the human makes the calls that touch the buyer.

SDRs are ads

My guest said something on air that reframed the whole debate for me. Twenty years ago, how did every company grow? Pour money into SDRs and buy a bunch of ads. Then he said it twice: "SDRs are ads." Those were your two ways to grow a business.

If an SDR is really a form of paid attention, then an AI SDR is just cheaper paid attention, and when everyone buys it, the price of attention goes up and the quality goes down. That's what's happening in the inbox.

His favorite alternative is partnerships. At a company he helped grow earlier in his career, the plan was to partner with mid-market HR and payroll companies for distribution, and it taught him that partnerships can be a real path to scale, not just a lifestyle-business strategy.

Mine is the show. It's a partnership with every guest: they get the spotlight, I get the conversation. Neither of us is buying attention. That's why I think the answer to "do AI SDRs work" matters less than the question underneath it: what are you offering that people actually want?

How to test one without burning your domain

If you want to try an AI SDR, do it carefully. A few rules we follow.

1. Never send from your main domain. Use separate, warmed sending domains so a bad week doesn't damage your real inbox.
2. Start with a small, high-fit list. Don't let a new tool loose on your whole database.
3. Read everything for the first two weeks. Every outbound email, every reply it drafts.
4. Check the reply loop before you scale. On a client call this week, before launching a family office campaign, we held the launch until we confirmed the AI replies were going out from the right sender identity. One wrong "from" and the whole conversation looks fake.
5. Cap the volume. AI makes it easy to send a lot. Mailboxes don't care how smart the email was if the volume spikes.
6. Measure meetings and deals, not emails sent.

So, do AI SDRs work?

As an assistant, yes. As a replacement, mostly no, and getting worse as every inbox fills up with the same robot. The edge has moved. It's no longer about who has AI. Everyone has AI. It's about who uses it to free up time for real conversations.

That's how we built our own system. AI finds the guests, writes the invitations and drafts every reply. Nothing goes out until the founder approves it with a swipe. And the thing being sent isn't a pitch at all. It's an invitation onto your show, which no amount of AI fatigue makes less welcome. Here's where AI and the founder each fit.

Frequently asked questions