The Trillion-Dollar Wake-Up Call: Make Your Agents More Capable or Become a Legacy Platform

In early February 2026, something unprecedented happened in enterprise software. Over $1 trillion in market value evaporated from the SaaS sector in what analysts are now calling the “SaaSpocalypse.” The damage has been staggering: ServiceNow has shed over half its value from its highs. Salesforce is down roughly 28% year-to-date. HubSpot has cratered nearly 70% from its highs. Atlassian has lost over 20% in a single week and is down 73% over the past twelve months. January 29 alone was the worst day for software stocks since the Covid crash — ServiceNow fell 11% despite beating earnings for the ninth straight quarter, and Microsoft shed $360 billion in market cap in a single session. These aren’t speculative startups — they’re the pillars of the modern enterprise stack.

The accelerant? Anthropic’s launch of Claude Opus 4.6 and its agentic capabilities — including agent teams, a million-token context window, and deep integrations into workplace tools. But let’s be clear: this wasn’t a single product announcement destroying a trillion dollars of value. This was the market finally pricing in a structural shift that has been building for two years. And the implications run far deeper than stock tickers.

The Real Threat Isn’t AI. It’s Who Owns the User.

The conventional narrative goes like this: AI agents will automate work, companies will need fewer seats, and per-seat SaaS revenue will collapse. That’s true, but it’s the shallow version. The deeper, more existential threat is about something far more fundamental — who owns the end-user relationship.

For the past two decades, SaaS companies have built their moats around a simple proposition: they are where work happens. Salesforce is where you manage customers. ServiceNow is where you handle IT operations. HubSpot is where you run marketing. The user opens the application, navigates its interface, and lives inside its world. Complex workflows running off this central system of record deepen the engagement. The application is the experience.

Foundational AI models are inverting this entirely.

With the maturation of protocols like MCP (Model Context Protocol) — now governed by the Agentic AI Foundation under the Linux Foundation, with backing from Anthropic, OpenAI, Google, Microsoft, and AWS — AI agents can connect directly to the APIs and data layers of any SaaS platform. The user no longer needs to open Salesforce to update a deal. They don’t need to navigate ServiceNow to file a ticket. They ask their agent, and it happens.

This is the shift the market is pricing in. SaaS platforms are being reduced from applications to data sources. They still hold valuable, structured, proprietary data — but the interface, the workflow, the user relationship? That’s migrating to the agent layer.

From Application to Data Source: The SaaS Demotion

Think about what it means when an AI agent becomes the primary interface for knowledge work. The user wakes up, opens Claude or ChatGPT, and says: “What happened with the Acme deal overnight? Draft a follow-up based on the last three touchpoints, schedule a meeting with their procurement lead, update my proposal and draft me a deck.”

That single prompt touches a CRM, an email platform, a calendar, a document/contract repository and maybe a presentation tool too. The agent orchestrates across all of them. None of those platforms “own” the interaction. They’re backends. Legacy data sources behind an intelligent orchestration layer.

This is exactly what Block articulated when they described MCP’s vision: the future centers on users navigating through one trusted agent rather than context-switching between fragmented experiences. MCP Apps are now even starting to let developers build interactive UIs — calendars, checkout flows, data visualizations — that render directly inside the agent conversation. The SaaS application doesn’t just lose the user’s attention. It loses the surface area where value is created and captured.

When Anthropic launched agent teams in Opus 4.6 — allowing multiple agents to split tasks, coordinate, and execute in parallel — they demonstrated that the “one agent, one tool” model is already obsolete. We’re moving toward agent ecosystems that treat the entire enterprise stack as a composable set of capabilities. In that world, the agent platform is the product. Everything else is data.

StackOne Agentic Connectivity

The Per-Seat Model Was Already Dying. This Accelerates It.

Jensen Huang called the panic “illogical,” arguing that AI will use and enhance existing software rather than replace it. He’s partially right — the software doesn’t disappear. But the business model built on top of it is under severe pressure.

If 10 AI agents can do the work of 100 sales reps, you don’t need 100 Salesforce seats. You need 10. That’s a 90% reduction in seat revenue for the same work output. And with hyperscalers on track to spend upwards of $600 billion in capex in 2026 — the majority of it on AI infrastructure — money that’s being redirected from enterprise software budgets — the financial squeeze is coming from both sides: fewer seats and smaller budgets.

Salesforce has responded with Agentforce. ServiceNow is building AI workflows. HubSpot is embedding assistants. But here’s the uncomfortable truth: bolting an AI layer onto a legacy application doesn’t solve the structural problem. It’s the equivalent of a taxi company launching an app while Uber rearchitects the entire transportation model. The agents these SaaS companies are building are trapped inside their own walls. They can do things within Salesforce or within ServiceNow — but they can’t reach out, connect, and orchestrate across the full landscape of tools their users actually depend on.

And that’s exactly the gap that determines who wins.

The Path Forward: Deep, External Agentic Connectivity

If SaaS companies want to avoid being reduced to passive data stores, they need to do something counterintuitive: stop trying to keep users inside their walled gardens and start making their agents the most connected, most capable, most context-rich orchestrators in their users’ workflows.

This means their agents need to do more than automate internal functions. They need to reach outward — connecting to the dozens or hundreds of other tools their customers use, pulling rich context from across the stack, and taking meaningful actions everywhere, not just within the boundaries of their own platform. An agent that can only update records in its own CRM is useful. An agent that can update the CRM, trigger a Slack notification, create a Jira ticket, schedule a calendar event, and draft a follow-up email — all in one coordinated flow — is indispensable. That’s the difference between a feature and a platform.

But building this kind of deep, cross-platform agentic connectivity is extraordinarily hard. There are over 250 major enterprise SaaS platforms, each with its own API design, authentication model, data schema, and rate limits. Normalizing across them, handling auth flows, maintaining reliability at scale — this is years of engineering work that most SaaS companies simply cannot afford to build from scratch, especially not at the speed the market now demands.

Why This Is an Agentic Connectivity Problem

This is precisely the problem StackOne was built to solve. As an AI-agentic connectivity platform, StackOne provides the integration infrastructure that lets AI agents — whether built by SaaS companies, enterprises, or AI-native startups — connect to and take action across the entire enterprise software ecosystem.

Agentic Connectivity isn’t a solved problem — it’s a very new space. Claude CoWork’s main connectivity paradigm is using vendor-issued MCP servers, including those it offers “out of the box,” but these still have significant limits. They’re typically built for local-client usage, not for usage across the enterprise by Agents used by non-technical users. They typically only expose a shallow subset of the potential actions available via the underlying API and are rarely optimized to minimize token consumption and to avoid the Agent’s context window becoming overwhelmed.

StackOne offers over 10,000 pre-configured actions across 200+ enterprise connectors, all of which are available via MCP but also A2A and compatible with frameworks like LangGraph, Vercel, Google ADK, and CrewAI out of the box — enabling you to build agents not for just a handful of power users but which can be rolled out across an enterprise user base of thousands of “ordinary” users.

For SaaS companies staring down the agentic shift, StackOne represents a strategic shortcut — a way to give their agents the deep external connectivity they need without spending years building and maintaining hundreds of individual integrations. Instead of competing with the foundation model layer on intelligence, they can compete on connectivity, context, and action breadth. They can make their agents the most useful orchestrators in their customers’ workflows, precisely because those agents can reach beyond their own platform’s walls.

The SaaS companies that will thrive in the agentic era won’t be the ones with the best chatbots bolted onto their existing products. They’ll be the ones whose agents can go anywhere, do anything, and bring the full context of their users’ work into every interaction. That requires infrastructure-grade connectivity at a scale no single SaaS company can build alone.

Whilst we’ve called out the big names everyone knows in terms of B2B SaaS platforms, this picture is arguably even more pertinent for the wider B2B SaaS landscape who don’t have the same tight grasp on their users as the likes of Salesforce, ServiceNow, and HubSpot, nor the engineering and financial firepower to try to address this problem space directly themselves.

The trillion-dollar selloff wasn’t a panic. It was a repricing of reality. The question for every SaaS company now is simple: will you be the Agent, or will you be the API it calls?