The short answer
It’s the wrong fight. AI SDRs are better than people at speed, volume and memory. People are better at the one thing that still books the hardest meetings: a live conversation with a stranger. The teams getting results aren’t choosing. They’re splitting the job.
Put bluntly: AI is replacing the SDR who only sends email. It isn’t replacing the SDR who can work a phone.
Who should do what
Our read of the evidence below, job by job.
| Job | AI SDR | Human SDR | Who should own it |
|---|---|---|---|
| Replying to an inbound lead | Seconds, any hour | Minutes to hours | AI |
| Building and enriching the list | Fast, tireless | Slow, error-prone | AI |
| Account research brief | Good first draft | Better judgment, slower | AI, human checks |
| First-touch email at volume | Strong | Can't keep up | AI, human writes the core message |
| Follow-up sequences | Never forgets | Forgets | AI |
| The live cold call | Legally and socially limited | Strong when trained | Human |
| Handling an objection live | Weak | Strong when practiced | Human |
| Discovery and qualifying in conversation | Shallow | Strong | Human |
| Booking, confirming, logging notes | Strong | Tedious | AI |
Where AI SDRs win
Speed to lead
Harvard Business Review’s classic study of inbound leads (2011) found that companies who responded within an hour were nearly seven times as likely to qualify the lead as those who waited even an hour longer, and most companies were far slower than that. An agent answers in seconds, at 11pm on a Saturday. No human team competes with that.
Volume and follow-up
SaaStr’s Jason Lemkin predicts classic email-based SDRs will be “90% displaced within 12 months”, and says his own company now runs on about 1.2 salespeople plus 20 AI agents. Salesforce reports its own agents contacted 130,000 leads in four months and created 3,200 opportunities. Both are company-reported, not independent studies, but the direction is clear: the inbox and the follow-up belong to software now.
Where AI SDRs have struggled
Early results were oversold
In March 2025, TechCrunch reported that 11x, the best-known AI SDR startup, had listed companies as customers that had only run short trials. Former employees described churn of 70 to 80% of customers; 11x disputed this and said current retention was 79%. Whatever the exact number, many early buyers tried an AI SDR and left within a few months.
Adoption is still tiny
In the Bridge Group’s 2025 survey of 351 B2B companies, only 1% reported running AI SDRs.
More agents isn’t more output
Gartner predicts that by 2028 AI agents will outnumber sellers 10 to 1, yet fewer than 40% of sellers will say the agents improved their productivity.
Email is getting quieter
The channel AI SDRs live in is crowded. In Belkins’ 2025 data, cold email earned about 0.45% replies per email sent. Over the same year, 1,000 dials produced about 57 live conversations.
The live call is off-limits by default
The FCC ruled in 2024 that AI-generated voices count as artificial voices under the TCPA, which generally means prior consent. A cold call doesn’t have it. And buyers still want a person in the loop: 69% check what AI tells them with a sales rep (Gartner, 2026).
Human SDRs aren’t winning by default either
The Bridge Group’s 2025 report found only 60% of SDRs hitting quota, the lowest in the study’s history. Average ramp is 3 months, average tenure 1.9 years, and median on-target earnings $80,000. Meanwhile, 47% of Gen Z reps say they don’t get enough roleplay before customer calls (Salesforce, 2026).
An untrained human on the phone isn’t the answer to AI. A trained one is.
The model that works: split the job
Before the call
The orchestra
- Research brief
- List building and cleanup
- Inbound replies in seconds
- First drafts of openers and emails
The call
The solo
- The first four seconds
- The objection, handled calmly
- Discovery, in their words
- The ask
After the call
The orchestra
- Notes and CRM
- Follow-up sequence
- Confirming the meeting
- Call review for patterns
Who’s already doing it
Several well-known companies have said publicly how they split the work. Most of them sell AI, so read it as their own account, but none of them has taken the human out of the conversation.
HubSpot: agents find and warm the account, reps make the contact
CEO Yamini Rangan describes an “agent-first” go-to-market. Agents find and score accounts and handle 82% of inbound website chats on their own. HubSpot says it first assumed email sequences would do most of the prospecting work, and that this was wrong: “only a small percentage of meetings get booked through email alone.” Its prospecting agent was rebuilt to watch for buying signals and hand reps tasks at the right moment. The company reports more than 10,000 meetings a quarter from AI-personalized outreach, and a 13% higher win rate on deals where reps used AI guidance.
Intercom: the AI agent qualifies chat, SDRs work the phone
Intercom’s own AI agent, Fin, handles frontline inbound chat at any hour, qualifies visitors and routes the serious ones to people. It has its own pipeline target and its own owner, an “AI SDR program lead.” Intercom says its best month of chat pipeline came after the switch. Its human SDRs now focus on phone-based qualification, which Intercom says “still converts best,” and on building relationships across the buying group.
Vercel: the agent took inbound, the people moved to outbound
Vercel’s COO Jeanne DeWitt Grosser has described replacing a ten-person inbound SDR team with a lead agent built by one engineer. She said the agent was “as good as our humans were”, with lead-to-opportunity conversion holding flat. One SDR now reviews the agent’s work. The other nine weren’t let go: they moved to outbound prospecting.
Microsoft: Copilot for its own sellers
Microsoft says its sellers spend about 70% of their time on admin and other non-selling work, so it turned its own tools on its sales org first. In one business group, sellers using Copilot generated 9.4% more revenue per seller and had 20% higher close rates; one team saves about four hours a week each. Agents gather data and do the routine work; sellers stay responsible for the relationship and the decisions.
OpenAI: AI does the research, sellers do the outreach
OpenAI’s sales team uses AI research agents (built in Clay, running on its own models) to do the account research its best sellers used to do by hand: websites, LinkedIn, earnings reports. It says 100% of research tasks are now automated for its reps, and enrichment coverage rose from the low 40s to the high 80s percent. Sellers pull that research on demand inside Salesforce and use it in their outreach.
Salesforce: agents chase every lead, and it still hired salespeople
Marc Benioff has said Salesforce had more leads than its reps could call back, and that with agents “every lead can be followed up on.” It also set out to hire about 2,000 salespeople to sell its AI products, and kept hiring salespeople in 2025 while hiring no engineers, support staff or lawyers.
IBM: AI took back-office work, sales headcount grew
IBM CEO Arvind Krishna told the Wall Street Journal in 2025 that AI had replaced a few hundred back-office roles, yet IBM’s overall hiring went up, with more programmers and salespeople, the areas he said need “critical thinking.”
And a warning: Klarna
Klarna leaned hard into AI for customer service, then started hiring people again in 2025 after its CEO said cost had been too dominant a factor and quality suffered. That was support, not sales, but the lesson carries: automate the volume, not the relationship.
The pattern across all of them: software covers every lead and every signal; people work the conversations that matter.
How to decide for your team
- Lots of inbound, slow response? An AI agent on speed to lead is the easiest win in sales right now.
- Outbound that’s mostly email? AI will do it cheaper. Move your people to the phone.
- Complex deal, senior buyer, small list? Humans on the phone, with AI doing the prep. Every conversation is worth too much to automate.
- Thinking of an AI voice for outbound? Talk to counsel first, and test it on a small list before you trust a vendor’s numbers.
- Buying any AI SDR? Ask for references you call yourself, results after month three, and the exit clause in writing.
If you’re an SDR reading this
The part of your job that’s going away is the part you probably liked least: copying, pasting, sequencing. The part that’s left is the part that’s hard: picking up the phone and holding a stranger’s attention for thirty seconds. Get very good at that and you’re the rep every AI-first team needs.
Rehearse the 45 objections prospects actually say, out loud, with the answer hidden until you’ve said yours. Start Practice mode free, then grab the prompt library for everything around the call.
Questions people ask.
Will AI SDRs replace human SDRs?
Are AI SDRs worth it?
Can an AI SDR make cold calls?
What should human SDRs focus on now?
Sources
- Oldroyd, McElheran and Elkington, Harvard Business Review (March 2011). The Short Life of Online Sales Leads.
- SaaStr, Jason Lemkin (Jan 1, 2026). The real future of AI in sales.
- Salesforce (Feb 3, 2026). State of Sales, 4,050 sales professionals.
- HubSpot, Yamini Rangan (updated May 22, 2026). How we grow with agent-first GTM.
- Intercom, Pascaline Albin (Apr 28, 2026). AI is the answer to the sales growth-without-headcount problem.
- Tomasz Tunguz (Nov 30, 2025). Vercel's AI sales agents, from Jeanne DeWitt Grosser on Lenny's Podcast.
- Microsoft, Tracy Galloway (Jun 3, 2025). Embracing an AI-first strategy to revolutionize sales.
- Clay. OpenAI customer story (2026).
- Fortune (Jul 30, 2025). Marc Benioff interview on AI agents and Salesforce's own leads.
- Verdict (Dec 2024). Salesforce to hire about 2,000 people to sell its AI products.
- Gigazine, citing the Wall Street Journal (May 2025). IBM CEO: AI replaced hundreds of workers but created programming and sales jobs.
- Forbes (May 18, 2025). Klarna and human customer service.
- TechCrunch (Mar 24, 2025). a16z- and Benchmark-backed 11x has been claiming customers it doesn't have.
- The Bridge Group (Feb 6, 2025). 2025 SDR Metrics & Models Report (351 B2B companies).
- Gartner (Nov 18, 2025). By 2028 AI agents will outnumber sellers 10x, yet fewer than 40% of sellers will report improved productivity.
- Gartner (May 20, 2026). 69% of B2B buyers turn to sales reps to validate AI-generated insights (645 buyers).
- Belkins. Cold email response rates, 7.53 million emails in 2025.
- Belkins. B2B cold calling benchmarks, 175,000+ dials in 2025.
- Cooley (Feb 15, 2024). FCC: AI-generated voices are artificial under the TCPA.