The Numbers
SaaStr now runs on 3 humans and 20+ AI agents. Their Salesforce bill: up 83% year-over-year to $22,000, despite dropping from 10+ human seats to 2 human seats plus 1 API seat.
Their Notion bill: still paying, but usage dropped to zero months ago. Renewal decision pending.
Why This Matters for Sales Teams
The gap explains the pattern. AI agents use software they need to do their jobs. They ignore software they do not.
SaaStr's AI VP of Marketing and AI VP of Customer Success run on Salesforce. It holds contacts, pipeline, activity history. Four third-party AI GTM agents run win-back campaigns at 72% open rates through Salesforce. They query constantly, never stop writing, treat Salesforce as their central nervous system.
Usage up 100x. Salesforce captured that value through API calls, Agentforce, Data Cloud, and consumption pricing. Human seats down 80%, revenue up 83%.
Notion lost the same team without getting worse. Notion AI shipped strong features, re-accelerated growth. But the AI agents route around it. Meeting notes, wikis, project tracking: agents build custom real-time dashboards on top of Salesforce and Slack instead. No reason to log in.
The Rule
Is this software critical to AI agents being successful at their jobs? If yes, usage and spend go up. If no, usage and spend go to zero.
Most B2B software companies have not seen the revenue impact yet. The seats are still on the invoice. The usage already left. Renewals will tell the story over the next 24 months.
What This Means for ANZ Sales Orgs
If your team is piloting AI SDRs or AI account research, watch what they actually use. CRM usage will spike. Collaboration tools and documentation platforms risk becoming write-only databases.
Salesforce Australia has positioned Agentforce heavily in ANZ enterprise accounts. Pricing follows consumption, not seat count. That model works when agents drive 100x usage. Competitors stuck on per-seat pricing will struggle when headcount drops but workload does not.
The shift is not theoretical. It is showing up in software budgets now. Systems of record win. Everything else needs to prove why an AI agent would need it.