Claim
Seat-based SaaS companies growing under 5% annually are not in a temporary slowdown. They are in terminal structural decline. AI agents perform multi-seat work without requiring multi-seat licenses, eliminating the expansion tailwind that drove software annuities for two decades.
Mechanism
Traditional SaaS annuities compounded because customers added seats as their organizations grew. One AI agent doing the work of ten support reps means the customer never purchases those ten seats. Lemkin names this directly: "If an agent does the work of ten support reps, the customer does not need ten seats." Expansion stops being a tailwind and becomes a headwind.
Infrastructure companies avoid this fate because AI workloads consume their products at scale. Usage-based billing on compute, data transfer, and storage scales with AI adoption. Seat-based application vendors sit on the other side of the divide. Lemkin frames the bifurcation: "There is no longer one category called software. There is infrastructure that AI feeds, and there are applications that AI feeds on."
The second pressure is capital reallocation. Roughly $450 billion in hyperscaler infrastructure spend in 2026 is targeting AI. Every dollar flowing into AI infrastructure is a dollar that does not renew a seat-based SaaS contract.
Conditions
Holds when: the product's core workflow can be substituted by an AI agent; the pricing model depends on per-seat expansion; the company lacks a usage-based or outcome-based alternative.
Fails when: the product has a human-first or creativity-dependent workflow where agents do not substitute; or the company has already shifted to consumption or outcome-based pricing before the seat-count erosion begins.
Evidence
Lemkin anchors the framework with public market data from 2026:
- Dropbox full-year guidance projects revenue decline of 0.4% to 0.9%
- PagerDuty net dollar retention at 97%, signaling a contracting customer base
- Zoom enterprise net dollar expansion at 99%
- Software now trades at 22.7x forward earnings, below the S&P 500 for the first time
"A terminal state. Not slow growth. The end of growth, with a question mark over whether what's left is even safe."
"There is no longer one category called software. There is infrastructure that AI feeds, and there are applications that AI feeds on."
Signals
- Net dollar retention consistently below 100% across enterprise cohorts
- Agent-driven workflows replacing feature requests that would have driven seat expansion
- Competitive displacement coming from AI point solutions rather than rival SaaS vendors
- Infrastructure and consumption-priced alternatives growing in the same category
Counter-evidence
Some seat-based products have shown durable seat expansion despite AI agent proliferation because their workflows are human-first. Creative tools and relationship-intensive platforms have resisted the pattern. The bifurcation between infrastructure and applications is a useful diagnostic, not a clean binary. Several companies operate usage-based models alongside seats and may navigate the transition. The thesis may over-index on back-office and workflow-automation categories where substitution is more direct.
Cross-references
- 10 unaffiliated paying customers = PMF floor. Then T2D3 (or T3D3) maps the path to scale.: Lemkin's earlier framework on SaaS growth milestones predates the seat-model collapse he now describes. The T2D3 compounding assumption breaks when seat expansion reverses.