The consensus view is that AI-native software will make old SaaS obsolete.
I think the first-order effect may be the opposite.
Agents like Claude and Codex make existing systems more useful because they attack the weakest part of enterprise software: the interface. Looking back, the worst part about platforms like Salesforce or SAP was interacting with the UI. Entering data was painful, querying it was clunky, and stitching context across systems required human labor.
AI changes that.
Agents do the dirty work. They unify context and log it back into the system of record where it belongs. Then the organization benefits from higher-quality data. The old system becomes more valuable and the user escapes the clunky UI. The agent turns messy enterprise data into usable context. In Salesforce's case, this makes the true customer 360-degree view achievable.
This is why the “AI-native CRM replaces Salesforce” argument may be too simple.
A Fortune 1000 company does not casually rip out Salesforce or SAP for a Series C startup because the demo is better. Existing systems come from reliable vendors. A core value proposition people miss is enterprise security and permissions. Those in the business know that Salesforce's secret is its security structure. Initiating a switch from Salesforce would be a security and IT review nightmare. No CRO gets fired for keeping Salesforce. A lot of people could get fired for moving core customer or financial data into an unproven AI-native system too early.
That gives incumbents time.
It does not guarantee they win. The risk is that Claude, Codex, and agent platforms capture the value while legacy SaaS vendors become data utilities. If the agent is the interface, the model provider may get the incremental token spend and user mindshare. Salesforce or SAP may remain necessary without becoming faster growers.
The investment question is who captures the value when agents make legacy systems usable.
There are two paths for incumbents. The defensive path is that their systems remain sticky because they are the trusted data layer. The offensive path is that they build domain-specific agents customers pay for: sales forecasting, supply-chain planning, and service resolution.
This is why Agentforce and SAP Joule matter. Salesforce and SAP have the domain data, workflow depth, and enterprise permissioning to build agents that horizontal AI tools will struggle to replicate completely.
AI makes the interface less important. It makes the system of record more important.
The irony is that agents may do legacy SaaS a favor. They remove the UI tax. They fill in missing data. They make old databases searchable and useful. They give incumbents a second chance to become the operating layer for business workflows.
AI-native challengers will still win in many categories, especially where old systems are weakest or SMBs can move faster. But in the enterprise, the first wave of AI may not tear out the system of record. It may finally make the system of record work.