AI Agents Are Great at the Grind—But the Hard Part Is Still Human

Three out of every five B2B go-to-market teams we track are trying to automate the wrong half of the sales process. They burn months on AI SDRs that book meetings no one closes, while the truly broken stuff—the invoice chasing, the CRM decay, the follow-up sequences that die in drafts—sits there waiting for a script that never comes. Jason Lemkin at SaaStr just put a number to the split: 21 AI agents in production have closed millions at SaaStr, but none of them closed a deal that wasn't already leaning in. That gap is real. But it's narrower than most founders think—and the data we track on PainSignal shows it's moving.

Lemkin's piece is worth reading in full, but the thesis is simple: agents have eaten the SDR role, the follow-up layer, and the quote-to-cash paperwork. What they haven't done is the moment of judgment—concession authority, reading silence, navigating the political maze of a six-person buying committee. That's the AE residue. The easy half is gone; the hard half remains human.

Except the line isn't stable.

The rise of AI agents in sales is often framed as a story of top-of-funnel automation—SDR replacement, meeting booking, lead qualification. But our data shows the quiet revolution is happening in the back office. Of the 25,201 problems tracked on PainSignal across 89 industries, a startling number cluster around invoice collection, payment reminders, and CRM data hygiene. These are the tasks nobody wants to do, and they are precisely where AI is gaining ground first.

Take invoice collection. It's the kind of unglamorous, high-annoyance work that AI agents are already solving for early adopters. A missed payment reminder sequence is not a judgment problem; it's a workflow problem with a clear trigger. That's why agents can handle collections follow-ups with 7-day escalation, as Lemkin's team does, while still failing to negotiate a discount on a complex enterprise deal. The distinction isn't about intelligence—it's about ambiguity. Collections has none. Closing has all of it.

What's less discussed is that AI is already closing deals—just not the ones you'd put on a sales leaderboard. Think self-serve upsells, cart recovery nudges, renewal automations with preset guardrails. These are low-complexity, text-based transactions where the "buyer" is already in the product, already using it, and just needs a nudge or a small exception. In our data, we see problems around self-serve upgrades and win-back campaigns where users report successful AI-driven closure without any human touch. Lemkin would call these "PLG motions," not real AE work. Fair. But from a builder's perspective, it doesn't matter what you call it if it prints revenue.

The sharper point from our dataset is the velocity of the shift. The article cites Emergence's survey showing 36% of B2B companies cut SDR headcount while 28% grew AEs. That's the compression everyone sees. But the ICONIQ data he cites is even more telling: AI-forward companies at $10M–$25M ARR run about 20 total GTM FTEs versus 35 for their lower-adoption peers—43% leaner at the same revenue, with 67% quota attainment versus 59%. The teams that lean into AI don't just cut costs; they perform better. That's not a gradual adoption curve. That's a cliff.

For indie hackers, the implication is straightforward: the biggest opportunity right now is not building a better AI SDR. It's building the connective tissue between agents and the hard residue. Think about the "agent handoff" problem—when an AI conversation stalls, how does it get pushed to a human with full context intact? That handoff is where deals are lost, and it's a wide-open space for tooling. We track 30 problems in the Communication category alone, many of them around follow-up, qualification, and notification workflows. These are not exotic enterprise problems; they are daily annoyances that a small team could turn into a profitable product.

Investors should pay attention to a different signal. The missing AI AE is not a bug; it's a roadmap. The first company that figures out concession authority inside guardrails—not just deterministic discount caps, but dynamic, context-aware negotiation—will own the mid-market. Lemkin predicts that agents will genuinely close text-based deals within 24 months. Our data says it could be sooner. The patterns we see in self-serve upsell and win-back automation suggest that buyers are more willing to transact with agents than many founders assume, especially when the price is under a psychological threshold and the product is already being used.

But there's a trap here. The article's insight about the "easy half" is structurally correct: agents absorb classification, scheduling, and workflow tasks first. That means the value of the remaining human work goes up, not down. The AE who survives compression is the one who can read silence, handle a procurement curveball, and own a concession with accountability. If you're an agency dev serving B2B clients, the message is: stop pitching AI as a way to replace salespeople. Pitch it as a way to make your best AEs absurdly efficient—by stripping away the 70% of their job that doesn't require judgment. The data backs that story.

One more angle worth flagging: agent-to-agent transactions. Lemkin mentions Stripe's point about agents becoming buyers—catalog, policy, payment rail. That future doesn't need a human-grade closer at all. If that side gets solved first, the missing AI AE becomes even more conspicuous. We're already tracking problems around automated procurement and API-driven purchasing. It's early, but the pieces are moving.

The bottom line is that the boundary between what AI can and can't close is moving faster than the high-level industry narrative suggests. Lemkin's piece is right about the current state—agents are great at the grind, weak on judgment. But the grind is bigger than most founders realize, and the judgment gap is already shrinking in narrow domains. For builders, the money is in the seams: handoffs, guardrails, back-office automation, and agent-to-agent rails. For investors, the signal is in the quiet back-office wins that don't make headlines. The AI AE isn't here yet. But the parts are on the table.