How AI Disrupts CRM and What That Means for Your Stack

They're calling it the “SaaS Selloff”.

HubSpot is down 51%. Salesforce is the worst performer in the Dow. Klarna just canceled both Salesforce and Workday to build their own AI-powered stack.

The market is pricing in a thesis: AI agents don't need enterprise software. They just need a database with good APIs.

Is the market right?

My thesis: Partially. The category survives—systems of record don't die. But the pricing power erodes, and the market splits in two. Salesforce keeps the Fortune 500. Everyone else starts to leave. Disruption comes from the bottom, not the top.

This matters if you're a CRO evaluating your stack, a RevOps leader managing vendor relationships, or a GTM engineer deciding what to build versus buy. The answer will shape your technology decisions for the next five years.

Meanwhile, AI coding tools have collapsed the cost of building custom software. A GTM engineer can spin up a purpose-built system in a weekend. The build-vs-buy calculation that favored buying for decades is shifting.

If you know Clayton Christensen's work, this pattern might look familiar. Christensen was the Harvard professor who identified how market leaders get dethroned—not by superior products attacking their best customers, but by "good enough" products serving customers the incumbent ignores or overserves. His theory of disruptive innovation explains why Blockbuster lost to Netflix. Netflix started with DVD-by-mail for movie buffs who didn't need new releases on demand—a segment Blockbuster was happy to ignore. By the time Blockbuster noticed, Netflix had built the customer base and technology to launch streaming.

The same pattern might explain what's coming for CRM.

Satya Nadella lit the fuse in December 2024, declaring that traditional SaaS applications will "collapse" in the agentic era—they're just "CRUD databases with business logic," and once AI owns the logic, the apps become commoditized infrastructure. Worth noting: Microsoft has Dynamics 365, a CRM competing with Salesforce. But it's not core to Microsoft's business the way Azure and Office are. Maybe that makes it easier for Nadella to be objective about where this is heading.

The incumbents pushed back. Aaron Levie, CEO of Box, argues that AI agents actually increase the value of systems of record—they become the "traffic cops" governing what data agents can access and how workflows are controlled. Marc Benioff calls the "SaaS is dead" narrative "crazy talk."

Both sides have a point. But they're talking past each other.

Levie's argument is worth taking seriously. In a recent post on the future of enterprise software, he frames it clearly: software is the "codification of a company's processes"—the parts you don't want to change spontaneously. You buy software for non-differentiating activities you don't want to reinvent every quarter.

His analogy: software is like machinery in a factory; agents are like the people operating that machinery. Deterministic systems (software) handle things that need to happen the same way every time. Non-deterministic systems (agents) handle judgment calls. They're complementary, not competing.

When you have 100x more agents than humans hitting your systems, governance matters more, not less. Access control, semantic clarity on what "correct" data means, audit trails—these aren't trivial. Enterprise complexity, compliance requirements, and integration depth all favor platforms that have spent decades building connectors and earning certifications.

Here's the thing: Levie is defending the category, not necessarily the incumbents' current position. He even acknowledges the Christensen risk explicitly, noting that "incumbents tend to lose in markets when there's a new tech or business model innovation that is unattractive to them." He cites a piece on "context graphs" by Jaya Gupta and Ashu Garg, admitting that "many new categories of agent-native software will need to be created for the first time because no existing vendor logically can capture the way the agent needs to do its work."

That's a significant concession. Even the strongest incumbent defender is saying: systems of record survive, but some of today's vendors won't be the ones providing them.

The capabilities agents actually need are straightforward: access control for machines, semantic clarity on correct data, good APIs for integration, and governance. Do you need Salesforce's 25 years of UI development for that? Or could it be built fresh with modern architecture, optimized for agent access rather than human clicking?

Here's how the pattern works. Incumbents keep adding features and raising prices for enterprise. They overshoot what SMB and mid-market actually need. New entrants offer simpler, cheaper alternatives. Incumbents rationally cede the low end—lower margins, higher support costs, not worth the fight. New entrants improve and move upmarket. By the time incumbents notice, the new entrants own the talent pipeline and the narrative.

Salesforce did this to Siebel. Now AI-native alternatives may do it to Salesforce.

In the agent era, "good enough" looks different. SMBs don't need 25 years of UI refinement—agents don't use the UI. They don't need thousands of AppExchange integrations—they need clean APIs. They don't need enterprise compliance certifications—they need data portability. A simple database with good access control and agent-friendly APIs is good enough.

Enterprise stays locked in. Too much historical data to migrate, too many compliance requirements, too much integration complexity. Salesforce keeps the Fortune 500.

Everyone else starts to leave.

Jason Lemkin is running 20+ AI agents at SaaStr—one of the few teams with real production experience across both Agentforce and competitors. His verdict on the CRM's role: "When you have 20 agents working autonomously, they need a hub. Somewhere for their data to meet. Somewhere to resolve conflicts. Salesforce is that hub."

That validates the incumbent position. But then he says the quiet part out loud:

"The potential conflict? We pay more for those agents than we pay for Salesforce. Which is an existential question for Marc and his team. The agents are extracting the majority of the value."

A Salesforce advocate, running agents in production, is telling you: the CRM becomes the hub, but the value extraction shifts to the agent layer. This is why Salesforce has 2,000 people on Agentforce. It's not confidence—it's survival mode.

Strip the CRM down to its essentials and ask what you're actually paying for. Database? Commodity—Postgres, Supabase, PlanetScale. Access control? Solved problem. Integrations? Commoditizing fast, especially as agents learn to call APIs directly. The schema? Generic—accounts, contacts, opportunities. The UI? Irrelevant to agents.

What's left is historical data (which is yours, not theirs), compliance certifications (real but narrow), and the network of consultants and partners who know the platform—a declining asset as the next generation builds on simpler tools.

If CRM equals database plus access control plus integrations, and all three are commoditizing rapidly, then the valuation is built on switching costs and inertia—not differentiated value.

The market is splitting. Enterprise stays with incumbents for now—the switching costs are real. SMB and mid-market face different economics: less data to migrate, simpler needs, more price sensitivity. AI-native alternatives and custom builds become viable.

If you're evaluating your stack, a few things to consider:

Your leverage in CRM negotiations is increasing. The "where else would you go?" question has more answers than it did two years ago.

Data portability should be a top-three requirement in any contract. Bulk export functionality, API rate limits sufficient for agent development, post-termination access. If a vendor fights you on these, that tells you something about their confidence in winning on product versus winning on lock-in.

Start documenting your methodology explicitly. Your qualification framework, your stage definitions, your rules for what "correct" looks like—this becomes the semantic contract that agents need to operate. Whoever encodes that logic gains leverage.

The consensus is right: systems of record don't die. Agents need canonical sources of truth. Governance matters more, not less.

The consensus misses the point: the category survives, but the incumbents' pricing power doesn't—at least not downmarket.

The real question isn't whether CRM survives AI. It's what a database with access control and integrations is actually worth when those capabilities are commoditizing and your vendor's margins depend on lock-in.

When your next CRM contract comes up for renewal, you'll have more leverage than you did last time. The question is whether you'll use it.