SaaS Is Losing the Interaction Layer. What Comes Next?

In my previous piece on the SaaSpocalypse, I covered the market panic after Anthropic and OpenAI launched back-to-back agent platforms. Nearly $1 trillion in software stock value disappeared. The narrative was dramatic but simple: AI is coming for SaaS.

A month later, the dust has settled enough to ask the harder question. Not whether SaaS is dying, but what specifically is being taken away, and what is left when it is gone.

The answer centers on one concept: the interaction layer.

The Layer That Made SaaS Valuable

Every SaaS product has three layers.

1. The data layer. Where business records live. Customer profiles, transaction histories, compliance logs.
2. The logic layer. The rules and workflows that govern how data moves. Approval chains, pricing calculations, routing decisions.
3. The interaction layer. The interface humans use to see, manipulate, and act on data. Dashboards, forms, modals, onboarding flows, notification systems.

For two decades, the interaction layer was the product. It was the reason companies paid $30 to $300 per seat per month. Not for the data (they owned that already) or the logic (which was often basic). They paid because the interface made the data usable by humans at scale.

Satya Nadella said it plainly: business applications are "essentially CRUD databases with a bunch of business logic" and that the notion of SaaS apps as we know them will "collapse" in the agent era. He was not speculating. He was describing what Microsoft is actively doing with Copilot.

The interaction layer is where SaaS companies built their moats. It is also what AI agents are now making optional.

Agents Do Not Need Interfaces

When Claude Cowork runs a contract review, it does not open a dashboard. It calls an API, parses structured data, applies rules, and returns a result. When OpenAI's Frontier orchestrates a sales workflow, it does not click through a CRM. It reads from and writes to systems of record directly.

Agents need three things from software: an API, a data schema, and clear documentation. They do not need dropdown menus, onboarding tutorials, or "intuitive" navigation. The entire presentation layer that SaaS companies spent decades perfecting becomes irrelevant when the end user is not a person.

This is not a theoretical concern. Bain's 2025 Technology Report describes the emerging stack as three layers: systems of record at the base, agent operating systems in the middle, and outcome interfaces at the top. Traditional SaaS products occupy the bottom layer. The interaction and orchestration layers belong to AI.

Bessemer Venture Partners frames it more bluntly: the industry is shifting from "systems of record" to "systems of action," and "data without action is just expensive storage."

What Is Left When You Are Just the Database?

A product manager at a mid-stage SaaS company posted on Hacker News in February: "We're cooked. Our entire value prop was making complex data easy to interact with. If an agent can do that through our API, what are customers paying us for?"

The question is not rhetorical. It maps to real numbers.

Retool's 2026 Build vs. Buy Report found that 35% of enterprises have already replaced at least one SaaS tool with custom-built software. 78% plan to build more in 2026. And 60% of those custom builds happened outside IT oversight, what the industry calls "shadow IT."

The pattern is clearest with simpler tools. SaaStr noted that nobody is vibe-coding their own Salesforce. But they are replacing "$49/month SaaS tools that do 80% of what they need and 40% of what they don't." When the interaction layer can be generated on demand, charging a monthly subscription for a static one becomes a harder sell.

Software price-to-sales ratios have compressed from 9x to 6x, levels not seen since the mid-2010s. PitchBook reports that $25 billion in software loan volume was trading at distressed levels by end of January 2026, with 30% of all distressed debt in the loan market coming from software.

The financial markets are making a clear bet: the interaction layer alone is not worth what it used to be.

The Copilot Cannibalization Paradox

Here is what makes the current moment structurally different from past tech disruptions. The companies most threatened by the shift are the same ones trying to lead it.

Salesforce launched Agentforce in late 2024 and scaled it through 2025. It moved from "copilots" (assist humans) to "agents" (act autonomously). Then came the pricing problem. Salesforce introduced Agentic Work Units (AWUs), a usage-based model that charges per task completed. But every task an agent completes is a task a human user did not need to do, which means one fewer reason to pay for that human's seat license.

Microsoft embedded Copilot across Word, Excel, Teams, and its entire productivity suite. But it faces a similar tension in its business model. McKinsey describes a "Gen AI Paradox": more than 80% of companies report no material contribution to earnings from their gen AI initiatives. If Copilot remains underwhelming, customers will look to independent AI tools that promise clearer value. If it actually works and automates meaningful parts of knowledge work, companies may need fewer human seats. Either outcome puts pressure on the traditional seat-based model.

Every incumbent faces the same bind. Build agents that are too capable and they erode the need for human seats. Build agents that are not capable enough and customers will use agents from Anthropic or OpenAI instead. Salesforce, Microsoft, ServiceNow (which acquired Moveworks for $2.85 billion in 2025), and other companies attempting the same pivot face this same tension.

Tomasz Tunguz of Theory Ventures captured the difference: "In 2016, investors questioned valuations. In 2026, they question relevance."

Would Customers Buy AI From Their SaaS Vendor or From AI Labs?

This is the question that determines which companies survive in their current form. The data points in both directions.

The case for incumbents:

Avenir's January 2026 report found that 63% of enterprise buyers expect AI to help their existing software vendors, not replace them. Only 8% see those vendors losing out. Most companies are betting on their current tools to adapt, not on new ones to take over. PitchBook puts it simply: "replacing a core SaaS platform is effectively open-heart surgery for an enterprise." Oracle, Salesforce, and SAP benefit from decades of embedded data and habituated processes.

The case for independents:

History offers a counterpoint. In code generation, GitHub Copilot had every structural advantage: distribution through GitHub, Microsoft's backing, and a massive existing user base. Yet Cursor, an independent startup, captured significant market share by shipping faster with features like repo-level context and multi-file editing. In sales, AI-native startups like Clay and Actively hold 78% share by attacking workflows that Salesforce does not own.

Pricing is another pressure point. AI add-ons from incumbents carry steep premiums. Microsoft Copilot adds 60 to 70% on top of base subscription costs. Scaling from pilot to full deployment routinely reveals 500 to 1,000% cost underestimation. When independent AI tools offer comparable capability at a fraction of the cost, the math starts working against incumbents.

The pattern that emerges: enterprises will likely keep incumbents for core systems of record where switching costs are highest, but adopt independent AI tools for workflow execution where speed and cost matter more. The interaction layer, in other words, goes to whoever ships it best and cheapest.

The Uncomfortable Middle Ground

The reality is messier than either the "SaaS is dead" or "SaaS is fine" narratives suggest.

Too embedded to die. Enterprise software is deeply woven into how organizations operate. Compliance requirements, audit trails, integration dependencies, and institutional knowledge all create friction against replacement. Gartner predicts that 35% of point-product SaaS tools will be replaced by AI agents by 2030. That is significant, but it also means 65% survive in some form.

Too commoditized to charge what they used to. When the interaction layer is no longer your moat, you are competing on data and workflow logic alone. Those are defensible, but they do not justify $300/seat/month pricing. The per-seat model, which powered two decades of SaaS revenue growth, breaks when one user with AI agents does the work of five.

Cathy Gao of Sapphire Ventures argues the winning companies will look less like SaaS and more like "managed labor," with agents getting "job titles, budgets, and limits." Bain advises the same shift: "Stop charging for access and start charging for work done."

The companies in the strongest position share specific traits:

• Unique data structures that are hard to replicate (regulatory records, financial histories, healthcare data).
• Deep vertical expertise in regulated industries where compliance is non-negotiable.
• Network effects where the product gets better as more participants join.

The companies most at risk are the opposite:

• Thin products where the core value is a clean UI on top of generic logic.
• Horizontal tools that serve everyone but own no unique data.
• Per-seat pricing with no path to usage or outcome-based models.

Morgan Stanley's Katy Huberty calls the current situation "sentiment-driven dislocation," arguing that companies with deep systems of record hold essential proprietary data that AI needs to function. She may be right about the dislocation. But "AI needs our data" is a weaker position than "customers need our product."

What Are You Without the Interface?

This is the question every SaaS company should be asking right now. Not "how do we add AI to our product?" but "if AI handles all the interaction, what is our product actually worth?"

For companies whose answer is "we hold irreplaceable data and enforce critical business logic," the path forward is clear. Become the best possible system of record. Invest in APIs, data quality, and compliance infrastructure. Let agents be the interface and charge for the value underneath.

For companies whose answer is "we built a really nice dashboard," the path forward is less clear. A really nice dashboard is what AI generates on demand now.

The SaaSpocalypse narrative frames this as sudden. It is not. AI captured 61% of global VC funding in 2025, totaling $258.7 billion. The capital has been flowing toward this outcome for years. What changed in early 2026 is that the products arrived.

Claude Cowork showed that a single agent could replace workflows across legal, sales, finance, and HR. Frontier showed how an entire organization of agents could be orchestrated. Tines launched an "AI Interaction Layer" in January 2026 to unify agents, copilots, and workflows through a single orchestration interface. The pieces are assembling quickly.

Marc Benioff dismissed the SaaSpocalypse fears, arguing that AI depends on the secure, structured enterprise data that SaaS platforms provide. He is not wrong. But depending on someone else needing your data is a different business than being the product people choose to use every day.

The interaction layer is leaving. The question is not whether SaaS survives. It is what SaaS becomes when the layer that justified its pricing for twenty years belongs to someone else.