SaaS Isn't Dying. It's Being Demoted.

The CEOs are conceding the interface point

Two weeks ago, Notion announced its developer platform and the headline term in the launch post was "orchestration layer." From the announcement: "Now Notion is your orchestration layer: a Decagon ticket routes to your coding agent, which proposes a fix and loops in your team to approve." Claude, Codex, Cursor, and Decagon now show up as native participants inside Notion workspaces. The External Agents API lets companies pipe their own agents in too.

Notion is the most telling example because nothing they've built since 2018 was about agents. Their entire business model is UX adhesion: keep you in the workspace, get you creating content, compound on switching cost. The product wins by being the place humans go to work. Two weeks ago, they publicly committed to being the place agents go to work too. That's a tell.

It's also not isolated.

In March, Linear CEO Karri Saarinen published a piece titled "Issue tracking is dead" launching Linear Agent. The new framing: Linear is now a "context-capture layer" while agents handle the engineering work. The data they shipped with the announcement: agents installed in 75% of Linear enterprise workspaces, agent work volume up 5x in three months, 25% of new issues now created by agents rather than humans.

Drew Houston put it cleaner last year while pivoting Dropbox to Dash: "They don't want to own the file, but rather the interface to all of them." He stepped down as CEO this week. His successor inherits the bet.

From the AI-first end of the market, the move shows up by default. Basecamp announced an "agent first, agent native" rebrand. Salesforce shipped Agentforce. HubSpot's CTO Dharmesh Shah built Agent.ai to 2M+ users with the explicit framing that "this is the decade of AI agents." And the next-generation challengers are starting from the agent layer rather than retrofitting. Twenty, the open-source CRM aimed at Salesforce, ships every Cloud workspace with a native Model Context Protocol (MCP) server.

This isn't a fringe position. It's the consensus across the people running the largest SaaS companies and the people building the next ones. That position and Kharazian's data can't both be load-bearing at the same time. If agents become the interface, "is SaaS surviving" stops being the right question. The right question is whether SaaS remains the interface layer or gets demoted to the capability layer behind one.

What "interface" actually means for value capture

When you log into Salesforce today, Salesforce is the interface. The interface is where value compounds. Network effects in UX, switching costs from muscle memory, brand premium for ease of use, distribution leverage from being where users start their workflow. All of it lives at the interface layer.

When an agent calls Salesforce on your behalf, Salesforce becomes the backend. The work still happens there. The data still sits there. The vendor still gets paid. But the user doesn't see it. The agent is the interface.

This is what Notion, Linear, Dropbox, Salesforce, and HubSpot are implicitly conceding. The locus of value capture is shifting, and they're betting on being the best place agents come to work rather than fighting to remain the interface.

Why Ramp's data doesn't catch this yet

Kharazian's critique is fair on the terms he's measuring. Seat counts, contract types, and vendor share are all real. But every metric he tracks measures the current arrangement, not the trajectory.

The transition from interface to backend doesn't show up as cancelled contracts. It shows up as flat growth in interface-tier pricing while agent-tier products eat the premium that used to attach to being "where work happens." It shows up in pricing power, not vendor logos.

His own data hints at it if you read sideways. High-intensity AI spenders saw token costs rise 13x year-over-year, and those firms are increasingly routing through OpenRouter for cheaper models. Comparing backends, optimizing for cost, treating providers as substitutable. That's the behavior agents will impose on the SaaS layer beneath them once the agent layer matures. Right now it's happening at the model layer. The application layer is next.

Where premium pricing actually compounds

There's one place where the demotion doesn't apply. Customers will pay premium pricing for context engines.

Data stores are already commoditizing. The differentiation between Postgres, Snowflake, BigQuery, and Mongo on storage and query is thin enough that the deciding factor in 2026 procurement is increasingly whether they connect cleanly into the agent platform the buyer uses. Vendors without clean connectors get cut from the shortlist regardless of how good the underlying tech is. The same compression is coming for SaaS backends. Once you become the backend, you compete on API quality, primitives, and integration depth. Most companies can't differentiate enough on those to defend premium pricing.

The exception is context. The systems that decide what an agent should see, when, in what compressed form, with what attribution will command premium pricing and strong loyalty. The reason is structural. Frontier labs (Anthropic, OpenAI, Google) make their money on inference and compute. They'll keep shipping better agent memory and better tools for compression. But owning context at the company-specific, workflow-specific, person-specific level isn't where their margin lives. They'll cede this territory because owning it doesn't pay them.

That cession is the opening. The companies that build real context engines — not retrieval with a fresh coat of paint but actual architecture that curates, synthesizes, consolidates, prioritizes, and stores intelligently — get to charge for it. They become the system every agent has to call before it can be useful. That's a stronger position than the SaaS interface ever was, because the value is structural to how AI products work rather than dependent on humans choosing to log in.

The actual decision facing SaaS leaders

The real question is which of three positions you're building toward.

Most SaaS companies will end up at the second path by default, because they can't do the first or third. The companies that recognize this early and price accordingly will be fine. The companies that keep pricing as if they're the interface will get repriced.