Support Is What Died

Cursor has more than 150 open roles. One of them is filed under “Customer Success,” and the description could have been written in 2016: “Manage a book of business with our top enterprise customers, owning the long-term success of each account,” lead rollouts “end-to-end: onboarding, enablement, internal champions, and scaled adoption,” and “drive renewals and expansion with AEs.” The title is AI Deployment Manager. The company with the best AI developers can buy is hiring the department part 2 said was dying, and it is not alone.

This is the other half of the argument. Part 2 made the case that AI and the product are reabsorbing customer success at AI-native, product-led software companies. This piece argues that what got automated was support, that the relationship and value job got renamed rather than removed, and that the moment any of those companies sells to a large organization, the department comes back with a new title and a higher salary. If you sell something complex, into a buying committee, with services or hardware attached, this is the case for your company. I could be convinced of the counter argument, and part 2 is where I made it.

The reversals, in the executives’ own words

Klarna is the cleanest record because its CEO narrated both halves. In February 2024 the company said its AI assistant was “doing the equivalent work of 700 full-time agents” and projected a $40 million profit improvement for the year. In February 2025 Sebastian Siemiatkowski said, “I am of the opinion that AI can already do all of the jobs that we, as humans, do.” Three months later, in May 2025, he told Bloomberg: “As cost unfortunately seems to have been a too predominant evaluation factor when organizing this, what you end up having is lower quality.” And: “Really investing in the quality of the human support is the way of the future for us.” By June the framing was that human service “is always going to be a VIP thing.”

Two more from the same year. Commonwealth Bank of Australia reversed 45 call-centre cuts in August 2025 after its voice bot launched and call volumes went up, saying “we should have been more thorough in our assessment of the roles required.” Duolingo’s CEO, four months after an “AI-first” memo that said the company would “gradually stop using contractors to do work that AI can handle,” said of the memo, “This was on me. I did not give enough context.”

Be honest about what these show. Klarna’s headcount kept falling, from about 5,500 to roughly 3,000 over two years, and the AI assistant never went away. None of these companies concluded that automation failed at volume. They concluded that the layer customers hold against you is the one you can’t run on cost alone, and they put humans back on exactly that layer.

The gap between the announcement and the payroll

Gartner has been counting. In a February 2026 survey of 321 service leaders, “only 20% of customer service leaders have actually reduced agent staffing due to AI.” Its prediction: “By 2027, 50% of companies that cut customer service staff due to AI will rehire employees for similar roles under different job titles.” A June 2025 survey had 95% of service leaders planning to keep human agents. On the customer side, a Gartner survey of 5,728 people found 64% would prefer companies not use AI in customer service at all, and 53% would consider switching to a competitor over it. Five9’s 2024 survey of 4,000 consumers had 75% preferring a human for support.

Those are service numbers, and I’d be careful applying them straight to customer success. The point they make is narrower and holds: the gap between what gets announced about replacing humans and what shows up on payroll a year later is wide, and the rehiring happens under new titles. Keep that phrase. It’s the rest of this piece.

People buy from people. I’ve believed that through every system I’ve built, and the place it holds hardest is the large purchase where the buyer is staking a reputation on the vendor as much as the product. That buyer is not asking a product to own the relationship. They’re asking for a name.

The leanest companies are rebuilding it under new names

Go back to Cursor’s careers page. Beyond the deployment managers there are solutions architects in five regions, a Regional Director of AI Deployment, forward deployed engineers in three regions, and field engineers for healthcare, life sciences, and the public sector. OpenAI posts “Customer Success Manager, Enterprise” and “Manager, Customer Success, Strategics.” Anthropic posts “Customer Success Manager, Tech (Strategics)” and tells the hire to “build trusting, strategic relationships with key customer decision makers in complex, matrixed organizations” and to “develop and execute change management strategies to drive end-user adoption.”

Harvey, at roughly $190 million in ARR, employs about 180 “legal engineers,” former practicing lawyers with eight to ten years in practice who run pre-sales discovery, onboarding, adoption, and the renewal and expansion motion, at on-target earnings of $220,000 to $320,000. Glean crossed $200 million in ARR with more than 1,000 employees, a ratio closer to classic enterprise software than to any lean-team story. Lovable, the leanest company in part 2, is standing up a forward deployed engineering function because, in its own words, “our enterprise footprint is compounding fast.”

The best dataset on the lean-team argument says the same thing. Kyle Poyar’s job-post analysis found AI-native companies had cut support to 2.2% of go-to-market headcount, and in the same paragraph: “They are still investing in customer success.” Redpoint’s playbook for AI-native companies, published in March 2026, says customer success “becomes critical in AI-native because everything starts as a proof of concept requiring education.”

Jan Young, a customer success leadership coach, made the observation under part 1 that AI-native companies have more post-sales people for the revenue they handle, and that those people work earlier in the process. The number behind it is ICONIQ’s 2025 State of Software: post-sales is 55% of go-to-market headcount at high-growth AI companies, against 23% at traditional software, with forward-deployed postings up roughly 12x. The post-sales share went up. It moved earlier, into deployment and proof of concept, and lost the CSM title on the way.

Andrew Marks, who was at Scopus when Vantive built the first customer success team, pushed back under part 1 on my own 60/40 split, and his objection belongs here. The 60% that every company shares, onboarding, adoption, health, renewal, expansion, is common in name and not in definition. “Ask four companies what onboarded means, and you get four answers.” His line: “An agent can take on specified work. The sixty was standardized in vocabulary, not substance.” Which is the job the deployment titles describe, because somebody has to write down what onboarded means at this customer before anything can automate it, and the companies above are paying lawyers and engineers to do exactly that.

Where this gets tricky is that a probabilistic product is harder to deploy than a deterministic one, not easier. Your enterprise buyer of an AI tool is purchasing something that behaves differently in their environment than in the demo, that needs their data and their workflows to show value, and that their security team has questions about. Every one of those is a human conversation, and the AI-native companies are hiring the humans to have it. They just stopped calling them CSMs.

The money is where the humans are

Cursor has positive gross margins on its enterprise accounts and loses money on individual developers, per reporting in April 2026. The segment that gets a named human is the segment that pays. Bessemer’s 2025 report on the fastest AI companies, the same one that celebrates $1.13 million of revenue per employee, describes the cohort as having “fragile retention and thin margins.” Growth without retention is the leaky bucket part 1 was about, and the fastest companies in software have it.

McKinsey’s November 2025 work on net revenue retention put the top quartile of B2B software at 113% and the bottom at 98%, and found that companies with “the most sophisticated value realization and adoption journeys produce NRR around seven percentage points higher.” Only 3% of respondents had what it called best-in-class coverage models. ChurnZero’s 2025 study of 793 revenue leaders found teams with CSMs reporting 98% net retention against 90% without. Seven points of NRR on a $100 million base is $7 million a year, every year, and that is the number a CFO weighs against the support savings.

Look at what Salesforce did with the 4,000 support roles it cut. By its own account it “redeployed hundreds of support engineers” into success architect and renewals roles and is shifting “more resources from reactive support to proactive customer engagement.” Its EVP of customer success summarized the logic in four words: “Prevention is the best cure.” The company that automated support hardest moved the people into the proactive layer. Zapier’s CEO, whose renewal agent does “90 plus percent of the work,” keeps a human for the rest: “There’s a lot of risk today if you don’t do that last mile work.”

If everyone owns the customer

Part 2 argued that the whole company can own the customer now because the whole company can finally see the customer. The counter is older than customer success. If everyone owns the customer, no one does.

Vantive’s customer success team in 1996 came out of a fight over who owned the account, which is another way of saying diffuse ownership had already failed at a company whose product was customer visibility. When part 1 went up, LeeRon Yahalomi put the whole third act in one sentence in the comments: “The org chart changes, the accountability doesn’t.” In 2026, 35.4% of companies report zero shared KPIs between sales and customer success, per the Customer Success Collective. Shared context does not produce shared accountability on its own. Somebody still has to be the one whose quarter is bad when the customer leaves. Part 1 closed by asking who noticed when customer success picked up a quota. Andrew’s answer: the CSMs noticed immediately, and leadership took considerably longer. Accountability is always clear to the people carrying it.

And the product, carrying the relationship by itself, has already produced the failure mode. In April 2025 Cursor’s own support agent told a user that being logged out when switching devices was expected behavior under a new policy. No such policy existed. Users said they were cancelling, and the CEO went to Reddit to explain the agent had made it up. A company whose product is AI, whose customers are developers, running an AI agent on its own support queue, lost paying customers to a hallucinated policy. The named human exists for the day that happens to an enterprise account.

Here is where I’ve landed on my own question from part 1. Everyone sells, and I still believe it. The department is situational: a pure product-led company can run the proactive layer in the product and give the relationship to a seller, and plenty of them are. What is not situational is accountability for the outcome. Where I helped move customer success out of cost of goods sold and onto a revenue justification, the justification worked because one team’s number moved when the customer stayed. Take the team away and the number has to live somewhere, or it lives nowhere.

Where I could be wrong about this

Job postings are a lagging indicator. Cursor and Anthropic are hiring the people they needed last quarter, and the agents Gainsight ships this year may be the reason those postings close next year. Gartner’s rehiring predictions are about service agents, and I’ve leaned on them for a function they weren’t measuring. “Rehired under different job titles” could describe exactly the reabsorption part 2 argued for, with the department gone and the work spread across sales and engineering, which would make this piece a long way of agreeing with the last one. And the counterexamples I find most convincing, Harvey’s lawyers and Cursor’s deployment managers, are at companies selling to enterprises. For the long tail, the agents may simply be right.

Where the series lands, and where it doesn’t

Part 1 asked which customer success you mean, the organization, the discipline, or the philosophy, and here is how the three fare.

The organization is situational. At a product-led company selling to individuals and small teams, the proactive job lives in the product and the relationship lives with a seller when it matters at all, and the department can go. The moment the same company sells to a large organization, the department comes back under a title that doesn’t say customer success.

The discipline is being split. Monitoring health, nudging adoption, and flagging renewal risk are becoming product features. Owning a buying committee’s outcome through a messy deployment is becoming a hybrid role, part engineer and part account manager, that the AI-native companies pay more for than they ever paid a CSM.

The philosophy is the only one of the three that survives every version of this, and it’s the only one Marie Alexander said she was building in 1996. Whether your company needs a department to hold it depends on whether anyone’s number moves when the customer leaves.

If you’re in the department, the postings say what to invest in. Jan’s advice under part 1 was “double down on industry expertise and upskill on AI and tech aptitude,” and the market has already priced the first half: Harvey pays former practicing lawyers $220,000 to $320,000 to own the customer’s outcome. The second half is what a deployment manager title means.

So is AI the death of customer success? For the department at a pure software company, probably. For the job, no. For the idea, it’s the first time since 1996 the idea has had a chance to work without the department. Which of those describes your company is the question I can’t answer from here, and it’s the one I’d like you to argue with me about.

Which layer got automated at your company, and which layer got renamed?

When the agent gets it wrong on your biggest account, who picks up the phone, and did anyone decide that on purpose?

If the department went away tomorrow, whose number would move when a customer left? If the answer is nobody’s, you have your answer about whether you still need one.