The Login Was Never the Point

For most of the last decade, the conventional wisdom on SaaS went like this: data is infrastructure, distribution is everything, and the real prize is the workflow layer on top. The database was plumbing. The interface was the product. Customers stayed because switching hurt, not because the software was genuinely irreplaceable.

That model is under serious pressure. And the companies feeling it most are not the ones you would expect.

The moat people thought was real

SaaS lock-in was switching costs dressed up as user experience. Companies like Pipedrive or Xero were not winning because their product was beautiful. They were winning because leaving meant re-entering years of records, retraining a team, and reconciling broken integrations.

But that logic had a flaw. The actual thing making migration painful was not the interface. It was the data underneath it. The interface was just the tax you paid to access what you had already built.

AI is now separating those two things, and the results are uncomfortable for a large portion of the software market.

A new consumer for your data

What has changed is not that data became more valuable. It always was. What changed is that we now have a consumer for it at scale.

Large language models are most useful when grounded in specific, real-world context. When an AI agent reaches into a company's Pipedrive instance and pulls the last six months of deal history before a renewal conversation, the product is no longer the CRM. The CRM is the memory. The AI is the sales rep.

This is what the Model Context Protocol (MCP) makes possible: a standardized way for AI systems to query external data sources. The race to become the primary interface is effectively over. The interface is the AI. What remains to be won is the right to feed it.

Companies that hold durable data by category

• Customer interactions: Pipedrive, Zendesk
• Financial records: Xero, Sage Intacct
• HR and payroll history: Personio, SAP SuccessFactors
• Support and conversation logs: Intercom, Freshdesk

What Salesforce just admitted

At its developer conference in April 2026, Salesforce announced "Headless 360", a platform-level initiative that exposes every capability as an API, MCP tool, or CLI command, so AI agents can operate the entire system without a browser. The official framing was direct: you never need to log into Salesforce again.

Salesforce is betting, and the logic holds, that in a world where AI handles the interface, its leverage actually increases. Its data becomes an input to every workflow, not just the ones its own product supports.

The segment that cannot make that argument

Consider the mid-market of software: project management platforms, BI tools, training systems, workflow automation. These businesses built real revenue on the premise that they were solving coordination problems. What they were actually selling was a marginally better view of information that lived somewhere else.

Analysts estimate AI agents could replace up to 35% of point-product SaaS tools by 2030. That number is probably conservative for tools that never owned original data in the first place.

The categories with the most exposure

• Workflow automation: orchestrating tasks across systems they do not own
• Project and task tracking: coordination layers with no original record underneath
• BI and reporting: visualization on top of data that lives elsewhere
• LMS platforms: content delivery without proprietary learning data

An AI agent that can read a Notion workspace and produce a project update is not a productivity improvement for Notion. It is a substitute for Notion.

The pricing model problem

Seat-based models assume value scales with the number of people who log in. In a world where one AI agent queries ten systems on behalf of one person, that assumption collapses.

The February 2026 market correction erased roughly $285 billion in software valuations in a matter of days. Investors started pricing in what happens when AI agents make half the seats in a 50-seat contract redundant overnight. Vendors that survive are already moving away from seat counts and toward pricing on outcomes delivered.

What durable value actually looks like

• They own data that is expensive or impossible to reconstruct
• They hold longitudinal history, not just a current-state snapshot
• Their data becomes more useful as more AI agents query it
• They can credibly shift pricing toward outcomes rather than headcount

For those companies, AI is not a threat. It is a distribution channel. The dependency deepens precisely because the data now flows into more decisions than ever.

The question the rest of SaaS has to answer

The software industry spent twenty years optimizing for the wrong metric: engagement. Daily active users, time in app, notification clicks. All of it assumed that using the software was the goal. But using software was never the goal. Getting the outcome was.

LLMs have separated those two things. You can get the outcome, the drafted proposal, the summarized pipeline, the flagged risk, without ever opening the application that used to generate it. The companies closest to real data are discovering the boring part of their product was the actual product all along. The question for everyone else is what do you actually own?