How Agentic AI Compresses SaaS Unit Economics — and What $224B in BDC Exposure Means
The market frames AI's threat to SaaS as replacement: AI will make software obsolete. That's the wrong frame. The structural threat is compression: agentic AI reduces the number of human seats required to perform a given workflow, which compresses the unit economics of per-seat licensing without eliminating the software. The companies aren't going away. The revenue per enterprise customer is contracting. And $224 billion in BDC-held software debt is not priced for the difference.
Per-seat SaaS pricing is built on the assumption that software consumption scales with headcount. Agentic AI breaks that assumption. A workflow that required 20 licensed seats — because 20 humans performed discrete tasks in sequence — can now run with 5 human supervisors and 15 AI agents. The software is still there. The license count is not.
Klarna's 2024 disclosure that it replaced 700 customer service roles with AI — later walked back in 2025 as the company selectively re-hired human agents — revealed the counter-argument: AI replacement is incomplete, and humans retain optionality advantages. But the counter-argument confuses replacement with compression. Klarna didn't need 700 seats. It needed fewer. The per-seat SaaS vendors serving the customer service workflow lost revenue regardless of whether every seat was replaced.
Business Development Companies hold an estimated $224 billion in exposure to enterprise software companies — the largest single sector concentration in BDC portfolios. BDC underwriting models assume the revenue stability of recurring SaaS contracts as the primary credit support for leveraged capital structures. Per-seat compression that reduces net revenue retention below 100% — which is the threshold at which the growth story inverts — creates coverage ratio deterioration that the credit models have not stress-tested.
This is not the same as "AI will kill SaaS." It is more specific: the revenue models that supported the leverage ratios in vintage 2021–2023 software LBOs assumed NRR patterns that agentic AI will compress in specific workflow categories. The categories matter. The vintage matters. The leverage ratio matters.
SGA maps which software workflow categories face the highest near-term seat compression risk, identifies portfolio companies exposed to NRR deterioration at the next renewal cycle, and provides credit structure analysis for BDC-held software exposure. The intelligence question: which portfolio companies are renewing contracts in 2025–2026 into a market where their customers have AI reduction data.
satish@sarrattglobal.com