ServiceNow's ARR growth has slowed from 23% to 14%. White-collar job openings are down 15% year over year. In Citrini Research's 2028 scenario, Zendesk defaults on a $5 billion direct lending facility in September 2027 and its debt marks to 58 cents on the dollar. Each step is a different chapter of the same book.
This is the moat-collapse signature of the AI displacement cycle. It will not be unique to Zendesk. Zendesk is just the cleanest worked example.
The four-step pattern
The pattern moves in a predictable order, and the order matters because each step closes the door on the recovery path that worked in the previous downturn.
First, ARR growth slows. The stock multiple compresses, but management blames "elongated sales cycles" and the analyst desks accept it. ServiceNow's slide from 23% to 14% is the canonical version happening in real time.
Second, AI customer service eats the actual product. A model running locally on the customer's stack handles tier-1 and tier-2 tickets at a fraction of the per-seat cost. The SaaS vendor's "AI feature" is a wrapper around the same models the customer can call directly. Net seat counts start shrinking, not growing.
Third, the private equity owner (Hellman & Friedman in Zendesk's case, Permira in others) discovers that the LBO math assumed 8% to 12% ARR growth in perpetuity. Debt service was sized to that assumption. At 0% growth, the interest coverage ratio breaks. At negative growth, the equity is worth zero before the debt is impaired.
Fourth, the direct lending fund that wrote the unitranche loan takes the haircut. Citrini's scenario has Zendesk's paper marking to 58 cents in 2027. That mark flows up to the life insurer that bought the BDC's senior tranche, which flows up to the annuity holders who thought they owned investment-grade exposure.
The pattern is a moat collapse, not a recession. A recession lifts when demand recovers. A moat collapse does not lift, because the moat is gone.
Why this SaaS cycle ends differently
The 2008 SaaS cohort survived the financial crisis because the underlying value proposition (pay per seat, host in the cloud, escape the on-premise upgrade cycle) remained intact. Customers cut seats and renegotiated contracts, but they did not stop needing CRM, ITSM, or helpdesk software.
The AI displacement cycle is different in kind. It does not reduce demand for the function. It reduces demand for the vendor. When Daron Acemoglu's "so-so automation" framework applies to white-collar SaaS, the savings flow to whoever owns the model that performs the function. That is rarely the SaaS vendor and often the customer running an open-weight model on their own infrastructure.
This is the structural problem with the per-seat business model. If the seat is replaced by an agent, the seat does not pay license fees. If the function is replaced by a foundation-model API call, the per-seat margin disappears even if the customer technically keeps the contract. Zendesk's revenue is a tax on the existence of human customer service teams. AI agents do not pay that tax.
The PE thesis was that mature SaaS is a bond: predictable cash flows, low churn, high margin. That thesis was correct for fifteen years. It stops being correct the moment the seats themselves become optional.
The daisy chain pulls everything else down
The dangerous part is what sits behind the SaaS LBOs. Citrini's daisy chain runs through life insurer annuity portfolios, private credit funds (Blackstone, Apollo, Blue Owl, HPS), and ultimately the residential mortgage book that depends on stable high-income white-collar employment.
Each layer was underwritten on the assumption that white-collar productivity software is structurally durable. Each layer is implicitly long the same trade. When the SaaS cohort takes the first haircut, the marks flow upstream, and there is no diversification benefit, because every layer is the same bet expressed in different paper.
In 2008, loans were bad on day one and the rating agencies missed it. In Citrini's 2028, the loans were good on day one. The borrowers were employed, the SaaS contracts were paying, the BDC marks were par. The world changed after.
If your portfolio is long any layer of the white-collar productivity stack, are you sure you are not long the same trade four times in a row?