The early AI SDRs went after OUTBOUND and didn’t deliver. Now a second wave is building, and this time it’s INBOUND-focused. Clay’s move is the signal: from infrastructure to customer-facing agents, and a flag in the CMO’s office.

AI SDR was the first hot agent-native category in GTM.

The early (and over-hyped) AI SDRs focused on OUTBOUND. They went head-on at everyone’s hardest problem. EVERY B2B company wants outbound to “just work,” but it’s actually super hard. In 2017 I felt this viscerally when I did a CEO to SDR experiment.

The AI SDR vendors sold Ava, Alice and Katie. Digital workers to replace your human SDRs. They’d work 24/7 to research prospects, write personalized emails, handle replies and book meetings, and revenue would rain down from the heavens.

It was easy to understand. And easy to fund. 11x, Artisan, Landbase and Bounti announced $114M in nine weeks in the fall of 2024.

Fast forward to 2026. They’ve all pivoted.

So the first wave didn’t deliver.

But the AI SDR dream didn’t die entirely. In 2026 a second wave is building.

This time it’s INBOUND-focused. In May, Vercel COO Jeanne DeWitt Grosser got on stage at SaaStr Annual and showed everyone her internal AI SDR program. Vercel’s agent took inbound lead qualification from 10 people to ~1.25. Startups like 1mind and Docket offer agents that answer, qualify and demo. And now the incumbents are buying and building. Salesforce bought Qualified. HubSpot is buying Warmly.

It heated up this month. Clay previewed an inbound SDR agent at Sculpt. Harmony, a voice AI sales platform built by monday.com, launched. Clay’s move is especially interesting because of the strategic signal -- moving from infrastructure to customer-facing agents and planting a flag in the CMO’s office.

The idea is older than ChatGPT. Conversica could claim OG status. Founded in 2007, it raised $87M by 2018. It was ahead of its time, but it never clicked. Then came the ChatGPT moment in late 2022. LLMs made the AI SDR idea seem possible again. This led to the first wave of AI SDRs as we know them today.

The outbound AI SDR wave hit fast in 2023 and 2024. Plenty of funding, ambitious founders, big claims. Lots of hype.

11x raised $76M across three rounds, the last at a reported ~$350M valuation. When a16z led the Series B, it said Alice “already has a 3x higher response rate than human SDRs.”

Artisan raised $36.5M and put “Stop Hiring Humans” on billboards around San Francisco.

AiSDR’s pitch was “replace your SDR with AiSDR.” Alta raised a seed for its “AI revenue workforce.”

Then the results didn’t show up.

In March 2025, TechCrunch reported that 11x had claimed customers it didn’t have. 11x disputed the report. Two weeks later, Artisan’s CEO told TechCrunch that first-generation AI SDRs “get a pretty low response rate” and have “relatively high churn.” AiSDR’s founder says he sold an AI SDR to 1,153 companies and watched 80% of them get the same result as a failed agency, “but faster and cheaper.”

Most of the failure case studies started with cold outreach. At LaunchDarkly, duplicate CRM records sent prospecting emails to existing customers, for “nearly a 0% response rate.”

This year AI SDR vendors changed their story.

In March, 11x published “AI SDRs won’t replace your sales team.” Alta published “The End of AI SDRs” the same month. In August, Artisan retired “Stop Hiring Humans” and announced its first human BDR. Ava still automates outbound. The billboard changed faster than the product’s ambition.

Luna shut down its platform. Its co-founder blamed “high churn, a crowded market, and the crackdown on cold emailing.”

AiSDR pivoted the other way. In September it relaunched as Ami, betting a broader agent can deliver the original promise for smaller businesses.

The churn shows up in public. In May, BuildWitt’s CRO was the star of an 11x case study: “our best guy on their best day.” On September 29 he posted, “We fired our AI SDR agent last night.” He didn’t name the tool. By October 1, the 11x case study page was a 404.

Meanwhile, the AI companies kept hiring human SDRs. At least 79 of 197 AI and cloud leaders have openings this week. More on that below.

Inbound qualification is a different job. The second wave is building here.

A visitor has arrived or a lead has raised their hand. You can give the agent approved information, qualification rules, account ownership and a calendar. It can answer, qualify, book or hand the conversation to someone. That’s a much more controlled job than creating demand inside a cold enterprise account.

Many companies still struggle with lead follow-up. We all know you want to be timely and helpful, but many ignore their leads. Sixteen days before Sculpt, Clay published a test: its agents filled out 6,346 demo forms. About two-thirds never got a reply.

Inbound agents aren’t new. Qualified launched Piper in April 2024. But the proof points, capabilities (like high-quality voice agents) and investment (startups and incumbents) have ramped in the last couple quarters. Clay put an exclamation point on it at Sculpt.

We are seeing more case studies from teams building their own. Vercel’s agent came from a GTM engineer who shadowed its best SDR and encoded her work. It runs on about $5,000 a year in infrastructure and tokens. ElevenLabs says its agent qualifies inbound in a four-minute median, against two days for its human SDRs. Greenhouse credits Piper with $4M in first-year closed-won revenue.

We’ve also seen new startup and investor energy. 1mind launched with $40M last November and was named Drift’s successor in March. Docket launched AI Seller, a website agent that qualifies and demos. Spara raised a $15M seed.

The incumbents are pushing too. Salesforce bought Qualified at a $1.2B acquisition-date fair value, then sold Piper from its marketing keynote at Dreamforce. HubSpot agreed to buy Warmly in June.

This month, Harmony launched: voice agents for sales, built by monday.com on technology it bought from OneAI. On October 8, Clay previewed its “first agentic seller,” an Inbound SDR Agent. It qualifies visitors by voice, demos your product in a sandbox and books the right rep. The waitlist page says: “Convert inbound leads your reps aren’t working.”

It’s a waitlist today, and still a big signal. In outbound, Clay was the infrastructure. The GTM engineering movement built its outbound systems on it. In inbound, Clay is building the agent. That moves Clay up the stack to customer-facing agents. It also moves Clay closer to the CMO, since inbound qualification usually sits with marketing.

Clay also has a habit of flattening categories. It started with enrichment. At Sculpt 2025 it added a sequencer and website intent, which is part of why the signal-based sales & ABM categories have been crushed. I tracked the deals last month: Warmly to HubSpot, Common Room to Zoom, Trigify to HubSpot. This summer Clay Ads 2.0 went live. Each launch pulls a point tool’s job into the platform. Inbound SDR is their next target.

Here’s my bet today. Reply and tell me what you see differently.

It looks more like customer support, a space where agents are flying off the shelves. Sierra says it passed $200M ARR. Decagon says $100M. A support conversation gives an agent a bounded job, a clear outcome and a person to hand off to. Inbound has many of those ingredients. Cold outbound is an entirely different beast.

There will be many different groups of vendors competing here. Outbound vendors are pivoting. 11x relaunched last week, and its homepage pitch now starts with “Convert inbound demand.” Incumbents will invest heavily. Pure plays will keep pushing. Support players may come in too. Fin already sells inbound qualification at $9.99 a qualification, and Sierra follows up on loan applications people already started. There are lots of ways to slice this: PLG overflow, after-hours, voice, live demos, considered purchases. It’ll be a big battleground with multiple approaches.

Agents take the signals, research, prioritization, drafting and CRM updates. People start the conversation with a stranger, run the big accounts, and pick up the phone.

The hiring data backs up my thesis. On October 21 I analyzed every live SDR posting at 197 AI and cloud leaders. At least 79 are hiring. Of 262 individual-rep postings, 81% assign phone work. Dialing for dollars is alive and well. Vanta’s posting says “Cold calling is at the heart of this role.” One posting in five assigns AI duty. Base44 wants an SDR who can “orchestrate a fleet of AI agents that handle high-volume outbound.”

Ramp’s internal AI SDR program shows where outbound goes. It built its own AI SDR in 2021. Gene Lee says it drove about 30% of pipeline for years before returns plateaued. Ramp shut it down last December, kept the data infrastructure and put more into tools for its sellers.

Builders will keep working on AI SDR projects for both inbound and outbound. Coding agents plus better GTM infrastructure (data, context layers, APIs, CLIs) make a custom agent cheaper every quarter. Vercel took its agent to production in six weeks. In Instantly’s public AI-versus-human test, the AI lane booked more demos than the human by week five. It’s a small vendor test, but the cold job isn’t settled. Some teams will get there.

Zoom out, and the AI SDR market is a preview of how GTM Tech evolves for the agent era.

There are two edges. Ship an agent that delivers an outcome, or ship the infrastructure & tools that agents need to work. Everyone is racing to these edges. The middle, SaaS with some AI bolted on, is getting squeezed.

In the AI SDR category, the infrastructure underneath did better during the first wave. Clay passed $100M ARR and raised at $7.1B to support GTM Engineers. And we’ve seen GTM Developer Tools like Exa (now past $100M in annualized revenue) and Apify growing very fast as tools for agents.

You might think of it as “Cursor for GTM” or “Vercel for GTM.”

I don’t see any buyers getting excited about stuff in the middle.

The first wave tried to replace the outbound team. It didn’t deliver, and the vendors walked it back. The second wave is going after inbound workflows, which look a lot more like support. Clay is moving to both edges, the infrastructure and the agent, while the middle gets squeezed.

How does the SDR function look at your company in 2026? I read all replies.