AI Sales Agents: What They Actually Do (And What They Don't)
A founder I know — I'll call him Dave because that's not his name — fired three of his five SDRs in September and replaced them with an AI sales agent he'd seen at SaaStr. The vendor promised the agent would prospect, qualify, write personalized emails, handle objections over email, book meetings, and follow up. All of it, autonomously, at a fraction of the cost of a human. Dave was so pumped he wrote a LinkedIn post about it. "The future of sales is here" type energy.
Three months later he quietly re-hired two of the SDRs. The agent had booked meetings, technically. But the meetings were terrible. Prospects showed up confused about what they'd agreed to. The "personalized" outreach the agent generated read like it had been written by someone who'd skimmed a company's About page and regurgitated it back with exclamation marks. His close rate dropped to nearly zero.
We grabbed coffee in January and he looked like a man who'd tried to return a used car. "It technically did everything the sales rep told me it would do," he said, stirring his latte more aggressively than necessary. "But none of it worked well enough to actually produce revenue."
That story is the entire AI sales agent market right now, compressed into one bad quarter.
The Hype vs. What Actually Works
If you've been to any sales conference this year, you've seen the demo. AI agent researches prospect, writes email, sends email, responds to reply, books meeting. Audience gasps. Standing ovation. What they don't show you is that the demo used a hand-picked company where the AI had rich data and the "prospect" was basically set up to say yes. Try running that same agent against a list of 500 mid-market companies where half have no Crunchbase profile and the other half's contact data is six months stale. Different story.
In reality, an AI sales agent that tries to do the full sales cycle end-to-end fails at the parts that require judgment. Qualifying a prospect requires knowing your product's weird edge cases, understanding which objections are real versus which are stalling tactics, and reading about forty micro-signals in a conversation that nobody has figured out how to codify. When a prospect says "I need to think about it," is that "send me the case study for my industry, I'm building internal buy-in" or is that "I'm being polite, please go away"? Experienced reps know the difference instantly. The AI guesses wrong about half the time.
The AI is bad at those parts. It is going to be bad at those parts for a while.
Okay, so what IS the AI actually good at right now? I've been running various agents for eight months and here's my honest list:
- Research. Give it a company name and it'll pull back firmographic data, funding rounds, tech stack, job postings, and recent news. Two minutes of agent time replaces what used to be a two-hour deep dive for an SDR. This is where I've seen the biggest, most immediate ROI.
- Data enrichment. Cross-referencing contact data across sources. Filling in missing fields in your CRM. Flagging stale data. The kind of work that nobody wants to do and everyone puts off until pipeline review day.
- CRM babysitting. Deal stages, activity logging, syncing notes between Gong and Salesforce and Slack. My reps were spending — and I measured this — about 35% of their day on this stuff. Zero of it generated revenue. All of it can be automated.
- Call summaries. You finish a 45-minute Gong call and by the time you've grabbed water, there's a structured summary in Slack with action items, competitor mentions, pricing discussion highlights, and a draft follow-up email that references things the prospect actually said. This one genuinely surprised me with how useful it is.
None of that is glamorous. None of it makes for a good conference demo. But it is the work that consumes the majority of a sales team's time, and it is the work that an AI agent can do reliably right now.
The "AI SDR" Question
I get this question at literally every sales meetup I attend: "Will AI replace SDRs?"
No. But it's going to reshape what the job actually is.
Think about what an SDR's day looks like today. They spend maybe 20% of their time actually talking to prospects — on calls, responding to replies, having real conversations. The other 80% is list building, research, data entry, writing emails, logging activities in Salesforce, and context-switching between six different tools. The AI sales agent is replacing the 80%, not the 20%.
The reps who will struggle are the ones whose only skill is volume — sending hundreds of emails a day using templates they didn't write. An AI agent can do that faster and cheaper. But the reps who are good at the actual selling part — understanding a prospect's problem, connecting it to the product, navigating a deal through a buying committee — those reps become more valuable when you take the busy work off their plate.
I've seen this play out at a handful of companies already. They didn't fire their SDRs. They gave each SDR an AI agent that handles research and data entry, and then raised quotas because reps now had six hours a day for selling instead of two. The output per rep went up. The team size stayed the same or grew. The job changed, but the jobs didn't disappear.
What a Good AI Sales Agent Setup Looks Like
Don't buy an all-in-one "AI SDR" box and expect it to run on autopilot. That's the demo talking. What actually works is uglier and more practical: separate agents, each one handling a specific chunk of busy work, wired into your existing tools.
Here's what our setup looks like right now:
Pre-call: An agent runs a research pass on the prospect — company data, recent news, their Crunchbase profile, open job postings, any LinkedIn activity worth mentioning. My reps get a one-pager before every call. They used to walk in cold or spend 20 minutes Googling. Now they walk in knowing the prospect just hired a VP of Engineering and mentioned "data infrastructure" in a LinkedIn post two weeks ago.
Mid-call: Gong is recording. The agent listens for specific moments — someone mentions a competitor, a pricing objection comes up, next steps get discussed — and flags them.
Post-call: Summary hits Slack within five minutes. Follow-up email drafted. CRM updated. Tasks created for anything the rep promised to do. Before agents, our reps were spending 15-20 minutes after every call on admin. Now it's basically zero.
Between calls: The agent watches our target account list for changes. Somebody raised money? New CMO? Job posting for the exact role our product serves? The rep gets a ping with context attached, not just a raw alert they have to investigate.
The rep sells. The agent does everything around the selling. That split is the one that actually works in 2026.
Why Use an Agent
Sure, you could do all of this with a disciplined rep who opens ten browser tabs before every call. I've met maybe two reps in my career who actually did that consistently. Everyone else means to do the research but runs out of time by the third call of the day.
I did the math on our team last quarter. Pre-call research was eating about 2.5 hours per rep per day (20 minutes x 8 calls). Post-call admin was another 2 hours. That's 4.5 hours of a rep's day spent on work that generates exactly zero pipeline. Now an AI prospecting agent handles the research in a couple minutes per account. The outbound automation agent manages follow-up sequences. And our call summary agent posts Gong summaries to Slack faster than I can finish my coffee.
My reps got about three hours back per day. They spend those hours talking to people. The conversations got better because the prep actually happened. Our pipeline reviews stopped being arguments about whether Salesforce data was accurate, because it finally was.
That's the real story of AI sales agents. Not the LinkedIn fantasy about replacing your team. The boring, measurable reality of getting your people out of admin hell so they can do the job they were hired for.
Try These Agents
- AI Sales Prospecting -- Automate prospect research, firmographic data pulls, and pre-call briefings
- Outbound Sales Automation -- Build signal-driven outbound sequences with automated research and personalization
- Gong Call Summary to Slack -- Turn Gong call recordings into structured summaries with action items, posted directly to Slack