CHURN IS DEAD
The Day Your CSM Got Promoted to Babysitter
11 min read · Team Design
Archive note: This issue predates the evidence ledger introduced in August 2026. Treat uncited benchmarks and examples as editorial analysis, not independently verified findings.
For a while now, Jason Lemkin has been arguing the traditional CSM role is dead. On June 3rd his SaaStr write-up put hard numbers behind it.
The evidence is a thing called QB. An AI VP of Customer Success managing 150+ accounts end-to-end, on a stack of three humans and twenty-one agents. And the detail that stuck with everyone, from one of the earlier episodes: QB autonomously emailed 83 sponsors at 12:20am with fully customised, account-specific check-ins. No human in the loop. No human needed.
He's right about the work. He's wrong about the conclusion. And the gap between those two things is the most important career question in customer success right now.
Because here's what actually happens when QB lands in a real enterprise org. The role doesn't die. It splits. It splits into two jobs that almost nobody was hired to do, and most teams are about to sleepwalk into the worse half of the split while their leadership calls it strategic transformation.
Let me show you both halves. Then let me give you the number that tells you which one you're building.
The work QB actually does
Start with what's true. QB managing 150 SaaStr sponsor accounts is real and it's impressive. I'm not going to wave it away to protect a thesis.
But look at the relationship shape. SaaStr sponsorships are largely single-vendor, transactional, renewed annually by one buyer who already knows the value because they can count the leads. The "work" is timely, personalised, high-volume outbound. That is precisely the work an agent does better than a tired human at midnight.
Now put QB on a Fortune 500 account where the renewal is a board-level budget fight, the champion just left for a competitor, security flagged a data-residency issue that froze the expansion, and three internal stakeholders are quietly knifing each other over who owns the platform.
QB can email all of them at 12:20am. None of those emails move the renewal one inch.
That's the split. The administrative core of the CSM job, the part that was 60 to 70 percent of the week, is now machine work. What remains divides into two distinct human jobs. Call them the supervisor and the closer.
Most CSMs hired in the 2021 boom were hired for neither. They were hired to build relationships and "be proactive," which is exactly the layer the agent just absorbed.
If your CSM job could be done by an agent emailing 83 people at midnight, it was already a dead job. The agent didn't kill it. It made it visible.
Before: the generalist who did everything badly because there wasn't time
Picture the role as it actually ran for the last four years.
You carried 40 accounts. Your Monday was an inbox of escalations, a sales handoff with no notes, two fire drills in Slack, and the one account threatening to leave this quarter. You wrote check-in emails between meetings. You prepped QBR decks at 9pm. You updated the CRM on Friday so the forecast looked clean.
You called this proactive work and you were proud of it. The dashboard showed your outreach cadence in green.
Here's the quiet truth about that green dashboard. The proactive nudge got sent. Nobody acted on the insight inside it. The health alert fired and you saw it and you had no bandwidth to do anything except note that you'd seen it. "Proactive" was a measure of how many messages left your outbox, not how many problems you prevented.
That's the version of the job an agent replaces wholesale. And it should. It was theatre.
After: two jobs, and one of them is a trap
Now the agent runs the outbound, the CRM hygiene, the first-draft QBR, the renewal nudges. What's left for the human is supposed to be "the strategic part."
That phrase is doing a lot of dangerous work. Because "the strategic part" actually contains three things, and only two of them are jobs.
Curation work, the supervisor. Judging, correcting, and approving what the agent sends before it damages a relationship. The agent drafts 40 check-ins over the weekend. Three of them are wrong in ways only a human who knows the account can catch: one references a feature the customer just deprecated, one congratulates a champion who got laid off Friday, one cheerfully asks for a renewal conversation the week after a Sev-1 outage. The supervisor catches those three. The supervisor is the reason the other 37 don't erode trust.
Commercial work, the closer. Navigating the procurement freeze, the champion change, the security review, the multi-stakeholder budget fight. The politics. The negotiation. The crisis where the customer's own change-advisory board won't approve the fix the agent already flagged six weeks ago. No agent does this, because the bottleneck was never detecting the problem. The bottleneck is the human org on the customer's side, and software can't navigate a CAB.
And then the third thing, which masquerades as a job:
The Babysitter Trap. Supervising agent output with no authority to change anything. You read the 40 emails the agent sent. You can't unsend them. You can't rewrite the cadence. You can flag a bad one but the workflow already shipped it. You sit in front of a queue of decisions the machine already made and your role is to feel responsible for them.
That's not supervision. That's busywork you're now liable for.
What a real supervisor needs that the trap denies them
The difference between a supervisor and a babysitter is a single word: authority.
This isn't abstract. Look at how most CS automation platforms ship today. The dominant design pattern is human-flags-after-send. The agent runs its sequence, the human gets a notification, and the human's options are to acknowledge, to escalate, or to add a note. The send already happened. The customer already read it.
That is the babysitter pattern hard-coded into a product. The human is positioned as a reviewer of completed actions, not an approver of pending ones. You're accountable for output you couldn't prevent.
A supervisor role requires three powers the trap removes:
1. Halt power. The ability to stop a send before it goes, not apologise for it after. Approval gates on anything touching a named risk account.
2. Rewrite power. The ability to change the cadence, the segment logic, the message, when the human spots a pattern the agent can't. Not file a ticket with the ops team. Change it.
3. Override-with-consequence. When the supervisor corrects the agent, the correction trains the next run. If your overrides vanish into the void and the agent makes the same mistake next week, you're a babysitter with a fancier title.
When a vendor sells you "human in the loop," ask exactly where the loop is. If the human enters the loop after the action completes, you didn't buy supervision. You bought liability laundering.
The judgment ratio
Here's the number that cuts through all of it.
The judgment ratio is the share of a CSM's week spent making decisions an agent cannot make, versus babysitting decisions it already made.
Decisions an agent can't make: the procurement strategy on a frozen renewal, the read on whether a quiet champion is disengaged or just busy, the call to escalate to the customer's CIO, the judgment that this account needs a human on a plane and that one needs to be left alone. Supervisor and closer work.
Decisions it already made: reading the 40 emails it sent, acknowledging the health alerts it generated, watching the dashboard refresh. Babysitter work.
Divide the first by the total. That's your judgment ratio.
If it's above 60 percent, you've redesigned the job. The human is doing what only a human can do and the agent is carrying the rest.
If it's below 40 percent, you haven't built a CSM. You've built a babysitter, and you're paying senior-CSM money for someone to feel responsible for a machine's output. The org will eventually notice the cost and cut the role, and they'll be right to, because the role you built genuinely had no judgment in it.
The zone between 40 and 60 is where most teams will land this year. That's the warning band. It means you bolted agents onto a dying job instead of designing a new one.
Here is exactly what to count on Monday. Pull last week's calendars and activity logs for three CSMs. Take every block, every logged task, and tag it with one of three labels: *agent-couldn't*, *agent-already-did*, or *agent-could-have*. Sum the *agent-couldn't* minutes. Divide by total working minutes. That's the judgment ratio for that human, measured, not guessed.
Do it for your best CSM and your most junior one. The spread will tell you whether your seniority maps to judgment or just to tenure.
The Supervisor / Closer Split, as an org-design decision
If the role bifurcates, you have to staff for the bifurcation. Pretending one generalist does both is how you end up with a team of expensive babysitters.
Let me be honest about the evidence here. No enterprise CS org I can point to has formally re-titled its team into "supervisors" and "closers" yet. This is a prediction, not a press release. So let me make it falsifiable: within 18 months, the CS orgs that retain pricing power on their renewals will have explicitly separated agent-supervision headcount from commercial-ownership headcount, and the ones that didn't will have flat or declining net revenue retention and a CFO asking why CS still costs what it does. If that doesn't happen, I'm wrong and you can hold me to it.
Here's the shape I'd design toward.
The supervisor role. Pattern-literate, detail-obsessed, fluent in the product and the data. One supervisor can credibly oversee the agent output across a large book, because the agent is doing the volume. The span of control goes up, not down. Where a CSM carried 40 accounts of manual work, a supervisor oversees the agent's coverage of 200-plus, intervening only on the corrections. You need fewer supervisors than you had generalist CSMs. That's the headcount reduction leadership is actually chasing, and it's legitimate, as long as it's named honestly.
The closer role. Commercially fluent, politically aware, comfortable in a room with a CFO who wants to cut spend. The closer carries fewer accounts than the old generalist did, because each account demands more depth: the freeze, the champion change, the security review, the board-level renewal. This is a higher band than the supervisor. It overlaps with what good account executives and account managers do, and it should be comped closer to them.
The ratio between them is the strategic call. A transactional, high-volume book skews toward supervisors. An enterprise book of complex, high-stakes accounts skews toward closers. Most orgs will need both, in different proportions per segment, and the failure mode is hiring one generalist band and asking them to flex between two jobs that require different people.
Pay them differently. Title them differently. Measure them differently. The supervisor is measured on correction quality and the trust-cost of what shipped. The closer is measured on gross revenue retention and the outcomes they navigated. Do not put them both on the same NRR scorecard, because one of them barely touches NRR and the other one is held hostage by product pricing decisions they don't control.
What I'd do this week
1. Run the judgment-ratio audit on three CSMs. Pull last week's calendars, tag every block *agent-couldn't / agent-already-did / agent-could-have*, compute the ratio. Below 40 percent on your senior people means you have a babysitter problem hiding behind senior titles.
2. Audit your automation for the loop. For every agent action that touches a customer, find out whether the human can halt it before send or only react after. Every "after-only" workflow is a babysitter factory. Demand halt power, rewrite power, and override-with-consequence, or stop calling it supervision.
3. Sort your current team into supervisor and closer. Not as a fixed identity, but as a question: which one is this person actually good at? Most generalists lean clearly one way. The ones who lean toward neither are the conversation you've been avoiding.
4. Reprice the two roles before finance does it for you. Decide the headcount ratio per segment, the pay band for each, and the span of control for supervisors. If you don't bring this design to your CFO, the CFO will bring you the Lemkin write-up and a number, and you'll be defending headcount instead of proposing a structure.
5. Stop comping closers on NRR. Put them on GRR, the floor they actually influence. Expansion is migrating to sales and to product upgrade prompts that convert better than any QBR. Comping CS on a number it half-controls is how you guarantee a blame conversation when the number moves for pricing reasons.
Lemkin showed a working agent doing the operational core of a CSM's job and called the role dead. He's right that the job he was looking at is dead. He was looking at the administrative CSM, and that one was already terminal. QB just turned off the life support and forced everyone to watch.
What he didn't show, because his accounts don't require it, is the closer navigating a procurement freeze or the supervisor halting a tone-deaf send to a grieving champion. Those jobs are not dead. They were always the actual job. The midnight emails were just the part we mistook for it.
So run the audit. Get your number. Then decide, deliberately, which job you're building, because the org chart is going to decide for you if you don't.
What's your judgment ratio?
By Kuber Sethi · All issues · Subscribe