The Great Unbundling: How AI Agents Are Picking Apart SaaS Suites

Table of Contents

• What "unbundling" actually means in the agent era
• Why SaaS suites bundled in the first place
• The seams where agents pry the suite apart
• Workflow layer vs. record layer
• The feature nobody loved but everybody paid for
• The economics: per-seat bundle vs. per-outcome agent
• Why the incumbents won't roll over
• A practical field guide for buyers and builders
• Insights Most People Overlook
• Frequently Asked Questions
• Conclusion
• References

What "unbundling" actually means in the agent era

Unbundling is an old story in software. A suite assembles a dozen capabilities, charges for the whole, and protects the price by making the parts inconvenient to buy separately. Then something changes the cost of buying the parts separately, and the bundle frays. Clayton Christensen's original framing of integration and modularity explained the pattern decades before anyone said "agent": value migrates to wherever the hard, non-good-enough work lives.

What's new is what gets unbundled. Earlier unbundlings split features apart, a point solution for expense reports, another for scheduling, a third for e-signatures. The buyer still operated the software. Agentic AI unbundles something deeper: the work itself. An agent doesn't replace the CRM screen with a nicer CRM screen. It reads the email, updates the record, drafts the follow-up, and books the meeting, the sequence of human actions the CRM existed to capture. When an agent absorbs the workflow, the application underneath stops being the thing you use and becomes the thing the agent uses. That demotion is the whole game, and it's the connective theme running through this cluster's pattern of agents doing the workflow the SaaS used to host.

So "unbundling" here has two layers. First, the familiar one: agents peel individual high-value tasks out of a suite. Second, the structural one: the agent inserts itself between the human and the software, and the suite's UI, the dashboards, the forms, the reports, quietly loses its reason to exist for that workflow. Hold both in your head, because incumbents are defending the first while the second is what actually reprices them.

Why SaaS suites bundled in the first place

You can't understand the unbundling without respecting why the bundle worked. Suites didn't win because buyers loved comprehensiveness. They won for unglamorous reasons that agents now partially neutralize.

Integration tax. Stitching ten point solutions together was expensive, fragile, and somebody's full-time job. A suite pre-integrated the parts. The data already flowed between modules. That convenience was worth real money, and it's why the middleware and integration layer became its own battleground.

Procurement friction. One vendor, one contract, one security review, one invoice. Enterprise buying is slow and political; the bundle made it survivable. Adding an eleventh point tool meant another vendor assessment that nobody wanted to run.

Switching costs and the data moat. Once your records, history, and configurations lived inside the suite, leaving meant migration pain. That gravity is real and durable, it's the basis of the argument that incumbents hold a data moat agents can't easily cross.

Single source of truth. The system of record had to be one place, or reconciliation became a nightmare. Suites promised that one place.

Notice that agents attack three of these four. They lower the integration tax by handling connections through APIs and even the UI itself. They reduce switching cost by sitting on top of whatever record system you have. They erode the single-vendor convenience by making it cheap to bolt a specialized agent onto an existing stack. What they don't easily defeat is the data moat, and that asymmetry shapes everything that follows.

The seams where agents pry the suite apart

Not all of the suite is equally exposed. The unbundling follows predictable fault lines.

Workflow layer vs. record layer

The cleanest way to predict what gets unbundled: separate the system of record from the system of action. The record layer stores state, who the customer is, what the contract says, what inventory exists. The action layer does things with that state. Historically a suite owned both, and the action layer's value got bundled into the record layer's subscription. Agents specialize in action. They're very good at the multi-step "look this up, decide, do the next thing" loop and comparatively indifferent to where the underlying data lives. This is exactly why the system-of-record vs. system-of-action distinction has become the strategic axis of the whole disruption debate.

In practice that means the agent is happy to let Salesforce or NetSuite keep being the database while it eats the labor that used to happen on top of that database. The incumbent keeps the moat and loses the margin on the workflow, a strange, uncomfortable middle ground that a lot of vendors haven't priced for yet.

The feature nobody loved but everybody paid for

Every suite carries modules that exist because someone needed them once, and they've been padding seat counts ever since. Reporting builders. Approval-routing engines. The "automation" tab that took a consultant two weeks to configure. These are prime unbundling targets because the value was always in the outcome (a report that's correct, an approval that routed correctly), never in the tool. An agent that produces the outcome directly makes the configured-feature obsolete. The disappearing dashboard, software that acts instead of displaying, is the sharpest version of this: if the agent acts, the screen that displayed the data for a human to act on has no job left.

Andreessen Horowitz's analysts have argued that AI is shifting spend from software seats toward agentic labor budgets, and this is the mechanism. The features most exposed are the ones whose value was always the labor they enabled, not the interface they presented.

The economics: per-seat bundle vs. per-outcome agent

Here's where the disruption gets teeth. Suite pricing is overwhelmingly per-seat: you pay for access, whether or not anyone does meaningful work in the tool. That model assumes humans operate the software. Agents break the assumption from both ends.

First, agents shrink seat count. One agent supervised by one person can do the work that filled five logins. The suite's revenue was indexed to headcount in the tool, and that headcount is the metric now keeping SaaS investors up at night, the looming seat-count shrinkage that doesn't show up in churn dashboards until renewal.

Second, agents reprice the value from access to outcome. A per-outcome or per-task agent charges for resolved tickets, qualified leads, reconciled invoices, units the buyer's CFO can map directly to labor saved. McKinsey's research on the economic potential of generative AI frames this as a productivity shift, but the pricing consequence is what disrupts: when value is sold by outcome, the per-seat bundle looks like paying rent on an empty office.

The trap for incumbents is that their revenue recognition, their sales comp, and their investor narrative are all built on seats. Repricing to outcomes cannibalizes the exact number Wall Street watches. That's the incumbent's dilemma in one sentence: defend the seat and lose to an agent-native rival, or kill the seat and detonate your own multiple. Neither door is comfortable, which is why so many vendors are choosing the third, slower option, bolting an agent onto the existing suite and hoping the bundle holds.

Why the incumbents won't roll over

It would be lazy to declare SaaS dead. The "SaaS is dead" thesis deserves a real steelman and an honest rebuttal, and the rebuttal is strong on several fronts.

The data moat is real. Agents need trustworthy state to act on. The incumbent owns that state, the access controls, the audit history, and the integration plumbing. An agent that can't reliably read and write the system of record is a demo, not a deployment.

Distribution beats raw capability. Incumbents already sit inside the enterprise, past the security review, with a signed contract and an account team. A startup agent with better technology still has to win procurement from scratch. As the cluster's distribution-versus-capability argument lays out, "we have the customer already" is a brutal advantage that pure-play agent vendors routinely underestimate.

Bundled agents are coming from inside the house. The mega-vendors are shipping their own agents on top of their own data, Salesforce, ServiceNow, Microsoft, and SAP all want to be your agent and your record at once. If the incumbent's agent is "good enough" and already integrated, the unbundling startup loses its wedge.

The honest read: incumbents will lose margin on workflows faster than they lose customers. The suite survives as the system of record and the trusted data plane; the contested ground is the labor layer on top. That's a repricing event, not an extinction event, but repricing a business built on 80% gross margins and per-seat expansion is plenty disruptive on its own.

A practical field guide for buyers and builders

For buyers evaluating whether to unbundle a workflow from your suite:

• Map the workflow to record vs. action. If the value you're paying for is action (work performed), an agent may deliver it cheaper. If it's record (trusted state), keep the incumbent and put the agent on top.
• Price the outcome, not the license. Ask vendors what a resolved unit of work costs, and compare that to the fully loaded labor it replaces, not to the old software line item. This reframing, from software budget to labor budget, is the negotiation that actually moves numbers.
• Watch for the data-access chokepoint. Some incumbents are already gatekeeping the integrations agents need. Before you commit, confirm the agent can actually read and write your record system at the depth required.
• Don't unbundle the system of record casually. Replacing a point feature is low-risk. Ripping out the database your whole company reconciles against is not. Unbundle the labor first; leave the moat alone.

For builders shipping vertical agents: your wedge is a specific, high-value, action-heavy workflow where the outcome is measurable and the incumbent's feature is widely resented. Win that beachhead, prove the outcome economics, and let the vertical-agent-versus-vertical-SaaS contest play out on terrain you chose. Don't try to replace the suite. Replace the worst, most expensive job the suite was hired to do.

Insights Most People Overlook

The bundle's first casualty is expansion revenue, not the logo. Everyone watches for churn. The real damage shows up earlier and quieter: net revenue retention. Seats stop expanding because agents absorbed the marginal users before anyone cancels a contract. By the time logo churn appears, the growth story has already broken, and most dashboards won't flag it until a renewal lands flat.

Incumbents are unbundling themselves to preempt startups, and accidentally training buyers to think in outcomes. When Salesforce or ServiceNow sells you a per-resolution agent, they validate per-outcome pricing as legitimate. Once a buyer has bought one outcome-priced thing, the per-seat line items on the rest of the bill suddenly look negotiable. The defensive move plants the seed of the broader repricing.

The agent doesn't have to be better than the feature, just better than the configured feature. Most suite power lives in modules that require expensive configuration to deliver value, and most customers configure them badly or not at all. An agent competes not against the feature's theoretical ceiling but against its real-world, half-configured floor. That floor is low, which is why "good enough" agents win faster than capability benchmarks predict.

Unbundling is asymmetric: easy to take the workflow, hard to take the record. The startups winning are the ones who accepted this and built on top of incumbent data rather than trying to replace it. The ones losing are still pitching rip-and-replace. The durable position for the next few years is "agent on someone else's system of record", which, awkwardly for purists, keeps the incumbent alive while eating its margin.

"Agent-proof" usually means "regulated, relational, or record-of-truth." The suite categories least exposed aren't the ones with the best UI, they're the ones where the value is being the authoritative, audited source of state, or where a human relationship and accountability are legally required. If your suite's defensibility rests on the interface, you're exposed. If it rests on being the trusted ledger, you have time.

Frequently Asked Questions

Does unbundling mean SaaS companies are going to disappear? Mostly no, but their pricing logic is under real pressure. The likeliest outcome is that suites retain the system-of-record role and lose margin on the workflow labor that sat on top. That's a repricing and repositioning, not an extinction, though it's painful for businesses built on per-seat expansion.

Which SaaS features get unbundled first? Action-heavy, outcome-measurable workflows where the suite's feature was always a means to an end: ticket resolution, lead qualification, invoice reconciliation, report generation, approval routing. The "configured automation" modules are especially exposed because their real-world value was usually the labor, not the interface.

Can't incumbents just add their own agents and keep the bundle intact? They're trying, and it's their strongest move because they own the data and the customer relationship. The catch: selling outcome-priced agents legitimizes outcome pricing across the rest of the bill, and it cannibalizes the seat revenue investors track. The bundle can hold, but its price gets renegotiated from the inside.

What's the difference between a system of record and a system of action, and why does it matter here? The record stores trusted state (the customer, the contract, the inventory). The action does work with that state. Agents excel at action and are indifferent to where the record lives, so unbundling tends to pull the action layer out while leaving the record, and the incumbent's moat, in place.

Is per-outcome pricing always better for the buyer? Not automatically. Outcome pricing aligns cost with value beautifully when outcomes are clean and measurable, but it can get expensive at high volume and creates disputes over what counts as a "resolved" outcome. Model it against your real labor cost and your volume curve before assuming it's cheaper than a seat.

How do agents get access to the data they need if the incumbent controls it? Through APIs, partner integrations, and increasingly the UI itself via browser-driving agents. But incumbents can gatekeep, throttling, restricting, or pricing API access to slow agent encroachment. Confirming durable, deep data access is one of the most important due-diligence steps before unbundling a workflow.

Where does this fit in the bigger GaaS picture? Unbundling is one front in the broader shift from software-as-a-service to agents-as-a-service. It connects directly to agent reliability (an agent absorbing a workflow must be trustworthy enough to act unsupervised), agent economics (per-outcome pricing), and agent security (an agent with write access to your record system is a serious attack surface).

Conclusion

The great unbundling isn't a single event, it's a slow, structural repricing of what SaaS suites are actually worth once an agent, not a human, operates them. The bundle held together for solid reasons: integration convenience, procurement simplicity, switching costs, and a single source of truth. Agents neutralize three of those four, leaving the data moat as the incumbents' real and durable defense. The fault line runs between the system of record (which incumbents keep) and the system of action (which agents absorb), and the economic shock is the move from per-seat access to per-outcome labor.

For buyers, the discipline is to unbundle the workflow without recklessly unbundling the record, and to price against labor rather than licenses. For builders, the wedge is the specific, resented, action-heavy job, not the whole suite. And for incumbents, the choice is genuinely hard: defend the seat and risk the agent-native challenger, or reprice to outcomes and detonate the metric Wall Street loves. Most will try to do both at once, which is why the next few years of this GaaS cluster will be defined less by who has the best agent and more by who controls the trusted data the agents need to act on.

References

• Skate to Where the Money Will Be, Clayton Christensen, Harvard Business Review
• AI Will Transform Services Into Software, Andreessen Horowitz (a16z)
• The Economic Potential of Generative AI: The Next Productivity Frontier, McKinsey & Company