Analysis
Salesforce delivered its fiscal second-quarter results, per the company, and CNBC framed the quarter as Marc Benioff getting his mojo back as AI strength quieted skeptics:
- Revenue: $11.345 billion, up roughly 11% year over year
- Agentforce ARR: crossed $1.5 billion, growth above 240%
- cRPO (contracted, not yet recognized): $33.5 billion, up 14% in constant currency
The bear case this attacks
For two years the short thesis on application software has been the "SaaSpocalypse": if AI agents do the work, companies buy fewer seats, and seat-priced software deflates. Salesforce stock spent much of 2025 and early 2026 trading on that fear despite consistent revenue growth. $1.5 billion of ARR from an agent product that did not exist in 2024 is the first large-scale counter-evidence from an incumbent -- the argument that the company selling the seats also sells the agent that sits beside it.
What the numbers do and do not prove
cRPO growth of 14% against revenue growth of 11% says bookings are running ahead of recognized revenue, which is the healthy direction. But Agentforce ARR is roughly 3% of Salesforce's revenue base, and the company has a long history of bundling new products into renewals in ways that make attribution generous. The honest reading is that agents are additive today and unproven as a replacement for the seat model over a full renewal cycle.
The competitive frame
Microsoft reports Copilot inside a much larger Microsoft 365 line, making direct comparison impossible by design. ServiceNow has pushed agentic workflows into IT service management with similar enthusiasm. HubSpot and Zendesk face the sharper version of the question at the low end, where per-seat pricing is most exposed. Salesforce's answer -- Claudeforce, built with Anthropic and reaching general availability in September, with every seller demoing it by Dreamforce -- ties its agent story to a frontier lab rather than to in-house models.
The Claudeforce partnership is the strategic decision worth studying. Salesforce spent years building Einstein and its own AI stack; pairing with Anthropic for the flagship agent interface concedes that frontier model development is not a fight an application company should pick. Adobe, Intuit and ServiceNow have made versions of the same call. The trade is margin -- inference costs flow to the model vendor -- in exchange for shipping speed and capability the customer can feel.
There is a counterweight the bullish read skips. Salesforce's overall revenue growth of roughly 11% is the slowest in the company's public history, and $1.5 billion of Agentforce ARR is growing off a base that was near zero eighteen months ago; triple-digit percentage growth is arithmetically easy at that scale and gets hard fast. The company is also carrying a substantial buyback and has been disciplined on headcount, both of which flatter EPS in ways that are not the same thing as demand. Adjusted EPS of $5.90 against net income of $4.844 billion is a strong quarter; it is not evidence that agents have changed the growth rate of the core business.
The test is the FY28 renewal cohort: whether accounts that added Agentforce renew their seat counts flat or lower. That is the number that settles the SaaSpocalypse argument, and it will not be visible for another year.
