The CEO of Salesforce went on All-In and called the semantic layer the key to SaaS company survival. Here’s why he’s right, and what most SaaS companies are getting wrong.
Marc Benioff went on the All-In podcast last week and (at 40:15) sent a bold signal to every SaaS CEO:
AI “needs to have that semantic layer,” and without one, “it just cannot work well.”
He was talking about Salesforce. The same goes for every other SaaS company.
The SaaSpocalypse isn’t about growth slowing; it’s about fading confidence in the SaaS model. Salesforce, HubSpot, and Workday are still growing, but their revenue multiples have been cut in half or worse. Nor is the SaaSpocalypse about AI. It’s a verdict on the wrong response to it.
Benioff is arguing that SaaS companies must own their semantic layer to not just survive, but widen their lead.
How SaaS Leaders Are Misreading Their Own AI Problem
Most SaaS companies are bolting an LLM onto their app, wiring up MCP endpoints, and calling that an agent strategy. Underneath, the data model is the same mess it always was. Metrics are littered across internal databases, business logic is inaccessible to customers, and buried in application code. The LLM, asked to reason against all of it, guesses.
As Benioff says, the fix is a universal semantic layer.
Why the Semantic Layer Is the New SaaS Superpower
A semantic layer is the encoded fingerprint of a SaaS provider’s moat, accessible to every AI agent on top. It encodes, operationalizes, and amplifies your SaaS special sauce. For example,
- Salesforce’s semantics grok customer behavior: which signals predict a closed-won opportunity, what a healthy pipeline looks like, and what a churn-risk account smells like.
- ServiceTitan’s semantics operationalize how to dispatch HVAC technicians: which jobs convert, which technicians are productive, and how seasonal demand bends.
- Procore’s semantics amplify the construction expertise that civil engineers respect: cost codes, change orders, RFI cycles, and the project hierarchy from owner to subcontractor.
- Workday’s semantics encode the workforce dynamics HR leaders bet careers on: retention risk by role, comp benchmark drift, succession plan gaps.
That domain knowledge is every SaaS provider’s moat. For most, none of it is reachable by an AI agent in any reliable way. The new superpower is to encode your fingerprint in a universal semantic layer, so AI operates with your domain knowledge, not around it.
Compare Notes with Chris Lynch and Jay Schuren
The question of whether and how to own a universal semantic layer is now an executive-level strategic decision, not a data engineering one. Snowflake treated it that way. SaaS companies in all domains must do the same.
To help SaaS executives work through that decision, AtScale is hosting a short series of private 15-minute briefings titled “The New Semantic Layer Superpower for SaaS,” featuring Chris Lynch, CEO, and Jay Schuren, President of AtScale. The conversation covers why Snowflake chose to use AtScale rather than build one, what that decision implies for every other SaaS company facing the same AI pressure, and how to evaluate the move for your own product roadmap.
Executive-only, by request, and limited in number. Request a conversation here.
SHARE
Guide: How to Choose a Semantic Layer