I talked to Liam Collins, Head of Performance Marketing at Instantly.ai, and the conversation kept circling back to one uncomfortable question: does the SDR role still need to exist?
Marketing now controls or influences up to 80% of the buyer's journey, depending on which survey you trust. Buyers wait longer to talk to a human.
And the traditional SDR job, list building, data verification, manual research was already eating most of a rep's week before they ever picked up the phone.
Liam has watched that job get rebuilt from the inside at Instantly.ai, a company that runs its outbound engine with zero SDRs. I wanted to know what actually replaced them, and what, if anything, is still worth a human's time.
Where the sales process actually breaks
Liam's first answer surprised me a little. I expected him to point at bad reps or weak scripts. Instead, he pointed at inconsistency.
"Having consistency across the reps and being able to just flatline and make that performance predictable has just always been a challenge," he told me. Some reps have their process locked in. Others are still ramping up, or are just weaker performers. When marketing pours leads into a pipeline with that much variance sitting underneath it, results plateau no matter how good the top-of-funnel work is.
He traced a lot of that inconsistency back to two things: enablement and admin. Building a target account list used to take hours of manual research at the start of every quarter. And once a rep finally got on a call, updating the CRM afterward was, in Liam's words, "a real pain in the arse." Marketing loses visibility into what's working the moment CRM data goes stale, and you can't scale what you can't see.
The lead response problem, and why it's basically solved
This is the part of the conversation that stuck with me. Liam described response times of two to three days as normal in the recent past. Not rare. Normal. A lead comes in, sits in a queue, and by the time anyone replies, the intent that brought them there has cooled off.
Instantly.ai's approach cuts that to minutes. An AI SDR enriches the account the moment a lead arrives, reads the inbound message, and sends a personalized reply without a human touching it.
Liam framed the shift plainly: the old process "required human intervention," and the new one doesn't, which is exactly why it's consistent. Every lead gets the same fast, informed response regardless of which rep would have handled it manually.
The same logic applies to old opportunities sitting dormant in a pipeline. Reviving those used to depend on a rep's gut feeling about which stalled deal was worth a second look. Now Instantly.ai tracks signals directly a site revisit, an ad engagement, a reopened email and triggers a follow-up automatically. Liam's summary: success used to come down to the individual; now it comes down to the systems and signals doing the heavy lifting.
I pushed back a little here, because this felt like an old problem with an old solution. Scheduling tools that route qualified leads straight to a rep's calendar have existed for years. Liam agreed, but pointed to something more specific than the tools themselves: the technical lift to connect a CRM, an outbound platform, and an ad account used to require dev resources most teams didn't have on hand. MCPs changed that. Now you describe the workflow to an LLM, and it tells you how to wire it up, no engineering ticket required.
Instantly.ai's no-SDR model
Here's the part I wanted him to walk through in detail, because "we don't have SDRs" is a big claim to make casually.
At Instantly.ai, a brand enters its own URL, and the system reads the website to infer the ideal customer profile, then selects target accounts on its own. A human still reviews and approves that list. From there, the agent builds the outreach sequences, personalizes every email using account-level research, and writes copy modeled on how a person would actually write it. The rep's job, at that stage, is to hit send, or not even that.
Liam was direct about where the org chart actually sits now: "We actually don't have any SDRs that do outbound, right? We actually have a go-to-market engineering team that does outbound." Paid media and that engineering-flavored outbound function run in lockstep, so the handover from marketing to sales happens without the usual friction of two teams working off different lists.
The one moment a human sales rep enters the picture is when a lead has already raised their hand and asked to talk to someone. Everything before that point list building, sequencing, sending, following up on signals, runs without a rep touching it. That's a genuinely different shape than "SDRs, but with better tools."
Why the discovery call is disappearing
I asked Liam directly about AI avatars replacing discovery calls entirely, something Fireflies.ai has experimented with publicly. His answer went further than I expected. He doesn't think a company should build an AI avatar to run discovery calls, but he also doesn't think discovery calls, in their current form, need to exist much longer.
"I actually think that it's redundant. I think it's mostly been redundant for a long time," he said. His reasoning: buyers increasingly arrive with an LLM having already done the groundwork, comparing tools, checking pricing, reading reviews, before a human ever gets involved. The generic, early-stage qualifying questions a discovery call exists to ask are questions the buyer has already answered for themselves.
Where he still sees a human conversation surviving is anywhere the purchase carries real cost and real risk. Software buying decisions almost always run through a committee, a CEO who wants to know why, a CFO who owns the budget. Liam expects that dynamic to hold "at least for now," and he thinks the seller's role shifts toward helping a champion build the internal business case rather than walking a single buyer through a pitch from scratch.
I'm not fully sold that the call disappears so much as it changes shape. Buyers still show up wanting to validate a shortlist an LLM helped them build, and someone still needs to answer for the business case internally. Whether that counts as the death of discovery or just its mutation is a question I don't think either of us fully resolved on the call.
What's left for humans
The clearest example Liam gave of why a human still needs to be in the loop wasn't about trust or relationships. It was about conflicting goals.
He described working with a company where the CFO wanted to improve retention and NRR while the founder wanted to cut CAC. Handing it to an AI agent without context, that's a recipe for a system optimizing the wrong thing. "If you give the agent that task without the right context, it's going to do amazing at reducing your CAC, but you're going to get completely rubbish leads," Liam said. An agent will find the cheapest path to a target number every time, whether or not that number is the one that actually matters.
That's the shape of human value Liam kept returning to: not typing the fastest prompt, but knowing which goals actually matter and catching it when a system optimizes for the wrong one. He also flagged something less obvious, that LLM output tends to regress to the mean because it's trained on what already exists. Differentiated creative and judgment calls about what to say and to whom are still, for now, a human job.
Final thoughts
The question I came into this conversation with was whether the SDR role survives at all. I don't think that's quite the right question anymore. Instantly.ai already runs at scale without one, and Liam expects more companies to follow, not because SDRs did their jobs badly, but because the research and admin that used to justify the headcount can now run itself.
What doesn't go away is the need for someone to decide what the system should be optimizing for, and to notice when it's optimizing for the wrong thing.
Liam's closing point is the one I keep coming back to: for the first time, there's no excuse for sales and marketing to be misaligned. The data, the signals, and the account lists are the same for both teams now.
If they're still not working off the same playbook, that's a choice, not a limitation of the tools.
