The short version: most software companies don't sell software — they sell seats. A seat is a proxy for a human doing a workflow. AI agents do workflows without seats. When the proxy breaks, the pricing model breaks, and the revenue model breaks with it. The stock usually breaks last — which is exactly why the short side cares.
Per-seat pricing was a bet on headcount
For twenty years, SaaS pricing has been a tax on hiring. Your support team grows from 50 to 200 agents? Your helpdesk vendor's revenue quadruples without shipping a single new feature. Wall Street learned to model this as "net revenue retention" — the share of last year's revenue a cohort of customers pays this year. Best-in-class SaaS ran NRR of 120%+: customers didn't just stay, they expanded.
Every premium multiple in software is downstream of that number.
Agents invert the equation
An AI agent that resolves 70–80% of tier-1 support tickets doesn't reduce a customer's seat count by a few percent. It changes the slope. The support team that was going to grow from 200 to 300 seats instead shrinks to 60. The vendor's expansion revenue becomes contraction revenue — same product, same customer, same renewal conversation, completely different trajectory.
This is the part the bulls consistently miss: the software doesn't have to get worse to lose the revenue. The customer's headcount plan does.
The repricing math
Consider a stylized seat-based vendor:
| Metric | Pre-agent era | Post-agent era | |---|---|---| | Net revenue retention | 115% | 85% | | Revenue growth | +25% | −8% | | Forward revenue multiple | 12× | 3× | | Implied stock move | — | −70%+ |
Notice that the business is still alive in the right-hand column. It still has customers, gross margin, a brand. But a 12× multiple priced for compounding becomes a 3× multiple priced for decay — and the equity eats the difference. This is why "software is going to zero" is directionally right as a trade thesis even when it's hyperbole as a literal claim: the multiple goes to zero-growth pricing long before the revenue goes to zero.
What to watch, in order
- NRR trajectory, not level. 108% falling from 119% is a worse signal than a stable 102%. The derivative is the tell.
- Seat-count disclosures quietly disappearing. When a vendor stops reporting seats or "paid users," ask why.
- Pricing-model pivots announced as innovation. "We're excited to introduce usage-based pricing" is often the sound of a company negotiating against its own installed base.
- Customer-side job postings. Roles like "agent orchestration lead" or "AI operations manager" at a vendor's largest accounts are renewal risk, posted in public.
- The guidance language shift. "Macro headwinds" doing more work each quarter while AI goes unmentioned in the risk factors — then suddenly over-mentioned.
The desk's framing
We treat every seat-based software company as a candidate for one question: what fraction of the workflows behind those seats can an agent do today, and at what cost ratio? When the answer crosses roughly half at a 10-to-1 cost advantage, history says the renewal cohort two years out is already lost — the market just hasn't met it yet.
That gap between already lost and not yet priced is the entire short thesis.
Stock Shorter publishes research and commentary, not investment advice. Short selling involves unlimited risk of loss. Do your own research.