Newsletter · · Ashutosh Agarwal

Sales AI Books More Meetings While AI-Native Companies Struggle to Keep Customers - Vertical Spotlight: Sales & Customer Service - Week of September 22, 2026

Startups and venture newsletter for the week of September 22, 2026. The Sales & Customer Service vertical spotlight, where Nooks reported 3x meetings booked with existing headcount, Zuora and RBC analysts argued per-seat pricing is breaking, and a customer-success podcast cited 48% median net revenue retention for AI-native software companies.

Vertical Spotlight: Sales & Customer Service

Week of September 22, 2026: The Week AI Admitted It Can't Grow Revenue On Its Own

This week's founders say revenue agents are booking 3x the meetings. But the same batch of podcasts asked the harder questions: can you keep the customers you win, how do you price a rep who's suddenly 5x more productive, and are the layoffs quietly being reversed?

The Landscape

If you listened to one podcast in the sales-and-support world this week, the message was almost cheeky: pouring more AI into your go-to-market machine does not, by itself, grow revenue. That was literally the title of the GTMnow episode featuring Loreal Lynch, the new chief marketing officer of the sales-agent startup Nooks: "Why More AI Won't Grow Your Revenue (and What Will)."

The throughline across the week's episodes was a shift in how people talk about AI in sales and customer service. A year ago the pitch was automation and headcount cuts. This week, again and again, founders and operators reframed it as augmentation (AI doing the grind so humans can do the parts humans are good at) and then admitted the hard part isn't the AI at all. It's rebuilding the process, the pricing, and the customer relationship around it.

Three tensions ran underneath everything:

  • Augmentation, not replacement. The people actually shipping these tools were surprisingly consistent: if an agent makes a rep dramatically more productive, you hire more reps, not fewer. The genuinely futuristic idea this week was inversion: the AI starting to prompt the human, instead of the human prompting the AI.
  • A quiet retention crisis. While everyone celebrates how fast AI companies are growing, one customer-success podcast surfaced a jarring statistic: AI-native software companies keep customers far worse than ordinary software does. Winning the deal is having a party; keeping the customer is doing the dishes afterward.
  • The pricing model is breaking. If AI makes a salesperson five times more productive, you can't charge five times more for the same seat, but you have to capture that value somehow. Two separate shows this week argued the decades-old "per-seat" software price is quietly collapsing into a messier mix of subscription, usage, and outcome-based billing.

Below are the companies and the specific numbers people put on the table, followed by the one debate that was genuinely unresolved on-air.

Companies to Know

Nooks: "revenue agents" that book meetings, and a vision of AI prompting the rep

What it does: Nooks builds AI agents for sales teams. It started narrowly, with an AI-powered dialer for outbound calling, then expanded this year into a full multi-channel sales-engagement platform. The founders are three Stanford engineers, and the CEO is Dan.

The metric that matters: New CMO Loreal Lynch (previously CMO of the AI-writing company Jasper, and before that a leader at Salesforce, Tableau and Stripe) said the results Nooks sees across its customers are "3x meetings booked, 2x pipeline. And that's with the existing headcount. That's without scaling additional reps." Customers she named include Cursor, Replit, Deel and HubSpot, mostly fast-growing tech companies that adopt AI early.

Why it's interesting: Lynch's whole argument is that automation-as-job-replacement is the wrong frame. In her words: "if you have an agent that is able to make a rep 10x more productive, why would you get rid of reps? Why would you not just double the amount or triple the amount of reps?" She compared it to the "10x engineer": companies didn't stop hiring engineers when coding got faster, they hired more and shipped more.

The part worth remembering is where she thinks this goes next. Today a rep tells the AI what to do. Nooks, she says, is building agents that learn to persuade through a feedback loop, so that "it's no longer that the sales rep is prompting the AI, but it's going to get to the point where the AI is actually prompting the rep: do this, go call this person, we recommend that you do this after the meeting in order to help accelerate this deal." She expects "the scales to start to tip here in the next year or two."

She was also blunt that the tool is the easy part: "it might be easy to build an individual agent, but scaling that across an entire go-to-market organization is much more challenging," because the agents need shared infrastructure (email deliverability, calling, LinkedIn outreach) and, harder still, the company has to redesign its processes: "you can't just take a new AI native tool and apply it to your old process. You've got to rethink your process from the ground up."

(The GTMnow Podcast, "Why More AI Won't Grow Your Revenue (and What Will) | Loreal Lynch (CMO, Nooks)," September 15, 2026)

Dust: an "AE operating system" where four agents work overnight while you sleep

What it does: Dust lets sellers build their own AI agents and chain them into workflows. Nic Siegle, a founding account executive there, walked through the personal "operating system" he's built on top of it.

The concrete picture: Siegle described a feature called "pods," an isolated workspace combining "teams of humans, teams of agents, and computers." He runs about four agents every morning before he's awake. By the time he walks his dog, the system has generated a spoken-word "podcast" briefing of his day, prepped every meeting (calendar, company research, contact research, product-usage data), drafted follow-up emails, updated the CRM, and sorted what needs his attention into an Eisenhower urgent/important matrix.

The claim that lands: His three highest-value uses were meeting prep ("saves 10, 15 minutes per meeting times X amount of meetings"), automatic post-call work triggered off the call recording (analyze the call, draft the follow-up, update the CRM), and call coaching across his last 30 days of transcripts: "how am I talking about pricing, how am I objection handling... a great way to get better." His rule of thumb for what to automate: "can I describe what I want to happen... can I get 80–90% of the way there so I can make the judgment calls and polish at the end?"

(The Crew Podcast, "Building GTM Agents Live + The AI-Pilled AE Operating System w/ Nic Siegle, Founding AE @ Dust," September 18, 2026)

Excite (on Vendasta): an agency that sells "AI employees" and hasn't lost a client

What it does: Klint Rudolph runs Excite, a Denver marketing agency. For years he sold websites, SEO, pay-per-click and social. Now he deploys AI agents, built on Vendasta's platform, to local businesses, and he says the agency itself now runs on them.

The metric that matters: Nearly 200 local businesses run on his AI stack, and he reports 100% client retention. His pitch: "I've sold websites, SEO, pay-per-click, content, social for years. This was easily the easiest sale I've ever made. My clients don't leave. They came for one capability and they stayed for everything."

Why it's interesting: His flagship agent, "Sophia," acts as a chat and voice receptionist that books appointments, captures leads and answers customers across phone, chat, email and social, around the clock. His aha moment was mundane and telling: after turning Sophia on, he found voicemails from people asking to build a website at nine o'clock at night, calls his three-partner team, with no office manager, had been silently missing for years. He trains every deployment like onboarding a person, using a house formula he drilled into his whole team (RCO: Role, Context, Outcome) plus hard constraints ("no guarantee of pricing, no fake capabilities, no claiming AI replaces staff") and a tiered playbook for high-, mid- and low-intent leads. His framing to customers: "it's not a chatbot... it's a new employee that has been trained on everything I know about your business."

(Conquer Local Podcast, "831 | He Sold AI Employees to 200 Local Businesses (100% Client Retention) | Klint Rudolph," September 15, 2026)

Recharge's "cool-hunting" agent: finding your next big customer before competitors do

What it does: Alan, director of marketing at Recharge (and former head of marketing at Skio, which Recharge acquired), built an agent to solve a sales-development problem: the whole Shopify software ecosystem uses the same lead-signal tool, Store Leads, so everyone is chasing the same brands.

The metric that matters: Every Tuesday his "cool-hunting agent" scours newsletters, X accounts, podcasts and consumer-brand directories, writes a short dossier on why each brand could take off, and hands it to his BDRs. "This week, it found 46. We are actively working 29 of them."

Why it's interesting: His argument is that off-the-shelf signal tools are "directionally accurate for established brands" but miss the interesting early ones. His example: the brand Groons. Store Leads "would not have told you anything about Groons" in 2023, and a few months ago it sold for $1.2 billion. The agent is a bet that finding great customers earlier is a bigger edge than scoring the same known list faster. (The same episode featured a marketer at DigitalOcean who chained together roughly a dozen Claude "skills" into an editing agent that checks a draft against the brief, the style guide and search visibility before she does, a reminder that a lot of "AI agents" in GTM this week were homemade, not bought.)

(The Dave Gerhardt Show (from Exit Five), "How I Build that [Agent] - 7 Marketers Show AI Workflows That Save Hours," September 21, 2026)

Zuora: the company trying to bill for AI when the per-seat model breaks

What it does: Zuora sells the billing-and-monetization plumbing underneath software companies. Katherine Shealy laid out why pricing is the quiet crisis of the AI era.

The numbers that matter: Two years ago, most software was priced per seat. Over the next twelve months, Shealy says, the mix between usage-based and hybrid pricing is "completely inverted." The reason is margins: "there's about a 20% margin difference, or a 20-point margin difference, between" traditional per-license SaaS companies and today's AI companies, because more AI usage means more real cost. She noted 60% of companies now change their pricing more than once a year, and that some are running as many as eight pricing models at once while they experiment, which quadruples the downstream billing complexity.

The blunt line: "Traditional SaaS pricing models are completely breaking down whenever it comes to AI. The value is really hard to define... the cost structure is totally dynamic." Zuora is even turning its own metering on itself, tracking the LLM token cost of running its AI product down to the individual engineer, so it can answer the question its own investors keep asking: how are we actually making money on this?

(AWS for Software Companies Podcast, "Ep224: Billing the Unbillable: How Zuora Monetizes AI on Amazon AgentCore," September 15, 2026)

Lassie (Bark Office): AI customer service that settles a claim in six minutes

What it does: Lassie is a European pet-insurance startup (Sweden, Germany, France and, as of a month ago, the UK), but its story this week was really about AI customer service. Its in-house agent system, cheekily named "Bark Office," reads a photographed vet receipt line by line and decides what's covered.

The metrics that matter: About 65% of claims in Germany are paid out in six minutes, with a stated error margin on automated claims of below 2%, versus what the founder cited as a 5% human error rate in the industry. Anything blurry, high-value, or uncertain still routes to a human. On the retention side: more than 90% of customers use the app (25% daily) for health guidance, quizzes and activity tracking via a partner tracker, which the founder tied directly to "better loyalty, reduced churn and higher LTV."

Why it's interesting: It's a clean example of AI in customer service where speed is the product, but the founder was careful about the guardrails, because "a mistake can impact both the animal's care and the owner's finances." Customers get a line-by-line email explaining exactly what was and wasn't covered, and can always dispute.

(Tech Talks Daily, "Rebuilding Trust in Pet Insurance With AI and Lassie," September 18, 2026)

Quick hits from the week

  • A CRO who stopped hiring SDRs. On Selling Intelligence, Thurman Sneed said he has four sales-development reps and no plans to hire a fifth, not because AI replaced them, but because AI raised each one's capacity to build pipeline, freeing his account executives for bigger deals. He also described using AI to predict churn better than the dedicated tools he'd bought to do it, so a human can "make a phone call before they ever know" a customer is looking at a competitor. His caveat: it only works if the data is clean and, crucially, accessible. (Selling Intelligence (formerly Selling the Cloud), "Ep. 144 – Building Predictable Pipeline and Breaking Revenue Silos with Thurman Sneed - Part 2," September 16, 2026)
  • The 60-30-10 rule for AI pricing. RBC's software analysts sketched a mental model for where seat-based software is heading: roughly 60% subscription/platform fee (enterprises still want predictability), 30% consumption/usage, and 10% outcome-based. Their honest caveat on outcome pricing: the hard part is defining the outcome and settling "who is responsible for it: was it the AI, the software, or the human being?" (Strategic Alternatives, "How AI is reshaping software business models," September 16, 2026)

One Debate: Is the "AI replaces headcount" story quietly being reversed?

This was the genuinely unresolved argument of the week, and it cut in two directions on two different podcasts.

The contrarian case. Stephen Klein, founder of Curiouser AI and a Berkeley lecturer, argued flatly that "the AI job apocalypse is being quietly canceled." His claim: companies sold AI to executives on the promise of needing fewer humans, laid people off, and are now quietly rehiring, because the tools are, in his words, "extremely unreliable." His memorable line: "The idea that you can replace somebody with something that is that unreliable is insane. It literally is like giving a monkey a nail gun and saying, now I can build houses more efficiently." He says it's happening quietly precisely because it's embarrassing and the stock market won't reward a company for admitting it was wrong, and he sees the CFO, not the technologist, now taking control of AI rollouts: "it's where the truth of the balance sheet lies."

The builders' case. The founders actually shipping sales AI don't buy the doom, but, notably, they don't argue for replacement either. Nooks' Loreal Lynch made the augmentation case explicitly: a 10x-more-productive rep is a reason to hire more reps, not fewer, just as faster coding didn't shrink engineering teams. Thurman Sneed's four-SDR story is the same idea in miniature: not fewer people, but the same people doing far more.

Why it's unresolved, and the number that complicates both sides. Here's the twist that neither camp fully answered. On The Customer Success Pro podcast, the host cited retention data ChartMogul published on AI-native software companies: a median net revenue retention of 48%, against roughly 82% for B2B software overall. In plain terms, the fastest-growing category in software is also the one that can't seem to keep its customers. Her verdict: "acquisition is having a party, while retention is doing all the dishes." She argued the misses aren't a talent problem but a system problem: teams handed a retention number "without the levers to pull" (no authority over pricing or packaging), renewal processes that only wake up weeks before the deadline ("you're not negotiating at that point, you are receiving a verdict"), and NRR treated like the weather instead of "a sum of hundreds of small commercial decisions."

So the real question the week left open isn't just "will AI replace sales and support reps?" It's sharper than that: if AI is genuinely booking 3x the meetings and closing faster, but AI-native companies are retaining customers at barely half the industry rate, then the bottleneck was never headcount or lead volume at all. It's whether anyone can keep the customers the machine is so good at winning.

Next week: Insurance.