The Product Is Taking Customer Success Back

At the fastest-growing software companies in the world, the department that owns the customer after the sale is sales, or engineering, or the product itself. Customer success still exists at most of them. It has a new title and reports somewhere else. That is the argument of this piece, and the evidence for it is a year old at most.

Part 1 made the point that customer success was carved out of support and services to own two things nobody else owned: making sure the customer got the value they paid for, and holding the relationship. This piece argues that the first job is being automated into the product and the second is going back to the people who had it before 1996. It’s a case about one kind of company: AI-native, product-led, pure software. If you sell hardware with a services contract, or into a regulated buying committee with nine stakeholders, part 3 is yours, and it argues the other side. I’m still working on this idea. Push back where it’s thin.

The support layer went first, and the people it reported to barely noticed

The job I described in part 1, a named technical rep because customers kept re-explaining their environment to a third person, was a support failure before it was anything else. Information didn’t travel between humans. An agent with the whole account history in front of it doesn’t have that problem, and the numbers on what agents now resolve are no longer projections.

Fin, the agent that used to be Intercom’s product and is now the company’s name, reports an average resolution rate of 76% across more than 12,000 customers and about two million resolutions a week. The rate has climbed roughly a point a month since launch, from near 35%. Intercom’s own support team was projected to grow 25% the year Fin shipped. It stayed flat. Salesforce agreed to buy the company for $3.6 billion in June 2026, a year after Marc Benioff said of his own support organization, “I’ve reduced it from 9,000 heads to about 5,000, because I need less heads.” Salesforce’s own account has Agentforce handling 2.6 million conversations at a 63% resolution rate with satisfaction scores matching human agents. Gartner expects agentic AI to resolve 80% of common service issues without a human by 2029.

None of that is customer success.

It’s the floor customer success stood on. Support was the function CS was carved from, the escalation path CSMs relied on, and the team whose cost justified the CSM’s higher salary. When support goes from 9,000 people to 5,000 and satisfaction holds, every budget conversation about the proactive layer sitting above it changes.

The proactive half is next, and the vendor that named it is selling the automation

Here is where the argument stops being about tickets. The company that gave customer success its conference, its book, and its software now sells agents to do the CSM’s job.

In August 2025, Gainsight announced “Retention-as-a-Service” with a Renewal AI it called “the first autonomous agent in the Atlas family.” By May 2026 it had Risk and Expansion Analysts that “automatically surface churn signals and expansion opportunities across accounts, flagging what matters months before a human alone would,” and a managed service, Atlas, “where Gainsight owns your outcomes.” The customer success vendor will now be your customer success department. Its new CEO, Chuck Ganapathi, put the logic plainly: “The only way you’re going to be able to get to the long tail is through agents and AI agents.” The example was an Okta long tail of tens of thousands of customers worth roughly $100 million, a book no human team was ever going to cover.

Law 4 of the Gainsight book, “Relentlessly Monitor and Manage Customer Health,” is now a product feature at the major vendors in the category. ICONIQ’s 2026 survey of go-to-market organizations found a single human CSM paired with an AI CSM covering the work of about 20 human CSMs, and 41% of companies with more than half their CS staff using AI daily. SaaStr runs an AI customer success agent on its own customers and says it outworks most human CSMs its founder has hired.

The clearest evidence is a job posting. Lovable, the Stockholm company that went from launch to $400 million in annual recurring revenue in roughly 15 months, lists an “AI/Scaled Customer Success Lead.” The job owns activation nudges, health scoring built on workspace activity and credit usage, a 90 to 120 day early-warning system for at-risk renewals, and routing of usage spikes to sales. Read it twice. That is the CSM’s entire proactive job, described as a system to be built, and the team it reports into is Sales.

The founders of the fastest companies say the title is the problem

In June 2026, SaaStr put the post-sales leaders of Lovable, Harvey, and AssemblyAI on a stage. Ryan Seams of AssemblyAI, whose company had already moved its CSMs to technical account managers and then to forward deployed engineers, said, “The CSM title is a tax on your customer experience.” Lovable’s line was shorter: “Customers don’t want an org chart, they want an outcome.” The panel’s shared claim was that more than half of what CSMs do has no measurable relationship to whether the customer stays.

Watch where the job lands at the companies that have the most AI to work with. Cursor, at a $3 billion run rate, calls its customer success manager an “AI Deployment Manager.” ElevenLabs, at $500 million in ARR with 530 people, files its “Deployment Strategist” under Engineering and Product and tells the hire to “embed yourself deeply inside strategic customers” and “own delivery outcomes as if you were part of their team.” Lovable files scaled success under Sales.

Three companies, three homes for the same work, and none of them is a customer success department.

The economics stopped supporting a department

In practice, there’s real difficulty in defending a budget line for work the product now does. Your CSM’s health score used to be the reason the department existed. When the product computes it, your budget meeting is about what the humans add on top, and the numbers coming out of the fastest companies make that a hard meeting.

Lovable reached $400 million in ARR with 146 employees, $100 million of it added in a single month. Bessemer’s 2025 State of AI report puts its fastest cohort at $1.13 million of ARR per employee. Kyle Poyar’s analysis of job-post data in May 2026 found customer support at 2.2% of go-to-market headcount at AI-native companies, 67% below other B2B software companies, with support postings down 37% year over year. GitHub’s SVP of customer success, running a post-sales organization of more than 550 people, has said publicly that the traditional CS allocation of around 10% of revenue is tightening toward 7% because AI is doing the post-sales work. Gainsight’s own 2025 index found its customers spending a median of about 3% of revenue on CS against roughly 8% for non-customers, and reported “more decrease than increase in overall headcount” across the function.

The structural number sits underneath all of it: only 45.6% of companies still run a named CSM per account, per the Customer Success Collective’s 2026 survey. Nick Mehta said the CSM population grew about 20% in 2021 and 2022 and roughly zero in 2023. The headcount curve of the profession flattened the year the agents arrived, and nothing since has bent it back.

The org chart from before customer success is coming back

Chris Degnan, Snowflake’s chief revenue officer, said it on a podcast in November 2023, before the agents were good: “I don’t believe in customer success. Because what does customer success do in most companies?” He had inherited the function and removed it. His replacement was a professional services organization, more sales engineers, and account reps who carry renewal and growth quotas. Sales owns the relationship. Engineers own the technical relationship. Support owns the experience.

That is the arrangement every company had before Vantive invented the CSM in 1996. Snowflake rebuilt it with people. The AI-native companies are rebuilding it with people and agents, and Intercom’s co-founder Des Traynor described where it ends: “Customer-facing roles will converge,” with “an AI Agent for resolving all customer queries, and a team of AI-native customer experts.” Eoghan McCabe’s version was that Fin would stop being a service agent and become “the world’s best Customer Agent,” spanning lead qualification, onboarding, support, success, and upsell. One agent across the whole lifecycle, and a smaller team of humans around it who are not organized as a department.

What “everyone owns the customer” looks like when everyone can see the customer

Marie Alexander said in 1996 that customer success “was not a department for us. It was a culture.” Lincoln Murphy has argued for a decade that it is an operating philosophy that includes sales. Both were right about the goal and had no mechanism for it, which is why the department got built anyway. Everyone owning the customer failed in practice because nobody could see the whole customer. The seller saw the deal, support saw the ticket, the engineer saw the bug, and the handoffs between them were where my TAM job came from.

That is the part that changed. When the seller sees adoption and credit usage before the renewal call, the engineer sees which accounts hit the bug, and the support agent carries the full history into the conversation, the job of holding the context stops needing a person. Everyone sells, in the sense that everyone who touches the customer can now act on the whole picture. The department existed to hold the context and the relationship. The product holds the context. The relationship goes back to whoever is in the room, which at a product-led company is the product most of the time and a seller or an engineer when it matters.

I helped build customer success departments at more than one company because the handoffs kept failing. If the handoffs stop failing, the honest question is what the department is for.

Where I could be wrong about this

The sharpest objection arrived before this piece did, under part 1, from Andrew Marks, who was at Scopus when Vantive built the first customer success team. The 60% of the job I called the automatable half, he argues, is “the half that looks automatable because everyone uses the same words for it.” Onboarding, adoption, health, renewal, and expansion are labels nearly every company uses and almost none has written down. “An agent can take on specified work. The sixty was standardized in vocabulary, not substance.” If he’s right, the proactive layer doesn’t get automated until somebody specifies it, and specifying it looks a lot like the job this piece says is going away. I take that up on Thursday.

Almost every number in this piece comes from someone with a stake: vendors selling the automation, founders whose valuations depend on lean headcount, a conference whose audience is those founders. The claim that over half of CSM activity has no relationship to retention came from a panel, not a study, and I couldn’t find the data behind it. Lovable’s job posting is evidence both ways, since a company with no customer success wouldn’t be hiring a customer success lead. And the whole case is scoped to companies where the product can carry the relationship. Most of the people reading this don’t work at one. Part 3 is about them.

The questions I’m carrying into part 3

If the product can see adoption, nudge the user, and flag renewal risk 90 days out, what is the named human at your company for? Name the task, not the value.

“Customers don’t want an org chart, they want an outcome.” Which department at your company would you make the outcome’s owner if customer success didn’t exist, and would they take it?

Everyone owning the customer failed in 1995 because nobody could see the whole account. Can your sellers and engineers see it now, and if they can, what are they doing with it?

Part 3 argues that what died was support, that the relationship layer got renamed, and that the companies with the best AI in the world are hiring for it. It has the harder evidence. Bring yours.