The biggest mistake in the “SaaSpocalypse” thesis is treating all SaaS companies as if they have the same exposure to AI.

Salesforce’s latest earnings make me think that argument might be too simple.

Agentforce ARR is already above $1.5B and growing 240% YoY, while Salesforce itself is still growing revenue around 11%. So AI doesn’t seem to be replacing the platform. It’s becoming another thing the platform can charge for.

The part I find most interesting is the pricing shift.

Traditional SaaS basically charges for how many people use the software. But if AI agents are doing more of the work, companies can start charging based on usage or work completed instead.

So maybe the real divide isn’t AI vs SaaS. It’s platforms with deep customer data, workflows and integrations vs smaller tools that mostly sell a narrow feature.

The second group seems much easier for AI to commoditize.

A field with few recent values may still feed an old report, integration, routing rule, permission check, or quarterly workflow. Deleting it too quickly can create a silent failure; keeping every field forever makes the schema harder to understand and maintain. What evidence do you require before retiring one? A workable process seems to need an owner, usage checks across automations and exports, a deprecation period, a replacement mapping when definitions changed, and a recoverable archive of old values. How do you handle fields that are technically populated but no longer trusted?