Agents are not killing SaaS. They are changing what gets billed.
The best current evidence points to hybridization: seats plus credits, work units, AI attach, consumption, and workflow outcomes. Broad named-user compression remains a hypothesis, not a public-company conclusion.
The useful question is displacement versus hybridization.
The original trade idea was a civil war between AI-native consumption software and seat-based SaaS. The project room kept the interesting part, but rejected the simple version: public evidence currently favors AI attach, credits, and usage layers being added to incumbent platforms.
Still alive. Microsoft 365 Commercial seats grew 6% YoY in the public metric set reviewed.
Salesforce Agentforce and Microsoft Copilot Studio show software moving toward action units.
SNOW, DDOG, MDB and observability platforms fit agentic workloads, but AI-specific ARR is thin.
Systems with data, governance, action rights, and distribution are more defensible than wrappers.
Where monetization is changing.
| Mechanism | Examples | Evidence Read | Open Question |
|---|---|---|---|
| Explicit agentic units | Salesforce Agentforce, Agentic Work Units, Flex Credits | Best public proof that enterprise apps are billing action and usage, not only seats. | Do work units produce attractive gross margin and customer ROI? |
| AI attach without compression | Microsoft Copilot, Copilot Studio, Workday agents, Atlassian Rovo | Public evidence shows seat-plus-AI, not seat-minus-AI, as of the reviewed disclosures. | Will paid AI attach eventually substitute lower-value named users? |
| Data and consumption workloads | Snowflake Cortex, Datadog AI ops, MongoDB, observability | Structurally aligned because agents need governed data and monitoring. | How much AI usage converts into durable gross-profit dollars? |
| Outcome / deployment platforms | Palantir, systems of action, workflow orchestration | Strong growth where software directly touches decisions and operating workflows. | How repeatable is the deployment motion across customers? |
| Cyber platforms | PANW, CRWD, ZS, FTNT | AI agents expand attack surface and governance needs, but direct AI-security ARR remains partly opaque. | Does AI security become incremental spend or budget reallocation? |
IGV is too noisy to be the whole thesis.
The local holdings pull captured 110 equity holdings and 99.97% equity weight, with the top 10 at 61.88%. Those top weights include Oracle, Microsoft, Palo Alto, Palantir, CrowdStrike, Salesforce, AppLovin, ServiceNow, Adobe, and Cadence. That is not a pure seat-SaaS basket.
Best next evidence pull: refresh PANW, CRWD, ORCL, and ADBE after June reports; build a pricing matrix across seats, credits, work units, usage, and outcomes; and track AI attach against RPO, NRR, gross margin, and seat/user growth.