Will Agents Commoditize the Application Layer? A Clear-Eyed Look at What Survives

Table of Contents

• The Question Behind the Question
• What "The Application Layer" Actually Means
• Where Commoditization Is Already Happening
• What Agents Cannot Easily Commoditize
• The Value Migration: From Screens to Outcomes
• Pricing Power Is the Real Battleground
• Who Wins, Who Gets Squeezed
• A Practical Framework for Founders and Buyers
• Insights Most People Overlook
• References

The Question Behind the Question

When people ask whether agents will commoditize the application layer, they're usually asking something more pointed: is my software business about to lose its pricing power? The application layer, the layer of the stack where SaaS lives, between the raw infrastructure below and the human above, has been the most profitable real estate in technology for fifteen years. Seventy-percent gross margins, net revenue retention above 120%, and the comfortable assumption that once a tool is embedded, it stays embedded.

Agents threaten that assumption in a specific way. If an autonomous agent can read your CRM, draft the email, update the record, and schedule the follow-up without a human ever opening the CRM's interface, then a lot of what you were paying for, the interface, the workflow, the "experience", stops mattering. The screen was the product. Remove the need for the screen and you've removed a chunk of the value.

But "commoditize" is a strong word, and it gets thrown around loosely. Commoditization means a thing becomes interchangeable, undifferentiated, and priced toward marginal cost. That's a high bar. To judge whether agents clear it, you have to stop treating the application layer as a monolith.

What "The Application Layer" Actually Means

Strip a typical B2B SaaS product down and you find four distinct tiers stacked on top of each other:

• The data tier, the records, the schema, the system of record. Your customers, your invoices, your inventory.
• The business-logic tier, the rules that enforce what's valid. Tax calculations, approval thresholds, compliance gates.
• The workflow tier, the orchestration that moves work from step A to step B to a human's inbox.
• The interface tier, the dashboards, forms, buttons, and reports humans click through.

For most of SaaS history these four tiers were sold as one inseparable bundle, and priced as one. You couldn't buy Salesforce's data model without Salesforce's UI. That bundling is precisely what gave the application layer its margins. Agents attack the bundle at its weakest seams, and the seams are not all equally weak.

The interface tier is the most exposed. It's the part with the least durable moat, a well-designed table view is nice, but it's not defensible against a model that can act on the underlying data directly. The workflow tier is next: a lot of "workflow software" is really just a state machine with a pretty front end, and state machines are exactly what agents are good at running. The business-logic and data tiers are far stickier, for reasons that have nothing to do with how clever the agent is.

Where Commoditization Is Already Happening

Look at the categories getting hollowed out first and a pattern emerges: they're the ones where the software was a thin shell over a workflow a human could describe in plain English.

Robotic process automation is the clearest casualty. RPA's entire value proposition was "record the clicks a human does and replay them." An agent that understands intent doesn't need the brittle click-path; it reasons about the goal. The category is being absorbed almost in real time, which is why we treat RPA's death by agents as its own obituary elsewhere in this cluster. Low-code and no-code builders face a related squeeze: their pitch was "describe your app without engineers," and a capable agent is a better natural-language-to-outcome interface than any drag-and-drop canvas.

Workflow-automation connectors, the Zapier-style middle layer, sit in the blast radius too. Their genius was packaging integrations so non-engineers could chain apps together. But an agent with tool access can call those same APIs and decide the branching logic itself, no pre-built "Zap" required. Andreessen Horowitz has written at length about how AI is reshaping the application and tooling layers, and the consistent thread is that value flows away from the orchestration glue and toward whoever controls the underlying capability and the customer relationship.

What unites all these losers is that they never owned data anyone cared about and never held the authority to act in a regulated context. They were convenience layers. Convenience is exactly what a sufficiently good agent provides for free.

What Agents Cannot Easily Commoditize

Now the other side, because the obituaries are overstated.

First, the system of record. Agents are stateless reasoning engines; they need somewhere to read and write durable, consistent, audited truth. The database with your actual customers in it doesn't get commoditized by an agent, the agent depends on it. This is the data-moat argument, and it's the strongest reason the largest incumbents aren't panicking. As Gartner has noted in its analysis of how generative AI reshapes enterprise software, the durable asset is proprietary data and the governed access to it, not the application chrome on top.

Second, the right to act. In regulated workflows, moving money, prescribing, filing with a government, approving credit, there's an accountable entity behind every action. An agent can draft the wire transfer, but somebody licensed and insured has to be on the hook when it goes wrong. The software that holds that accountability, the audit trail, and the compliance certifications is not interchangeable. You can't swap it for a cheaper one because the value isn't the feature; it's the liability someone has assumed.

Third, deep domain-specific data models and integrations. A vertical SaaS product for, say, dental practices encodes years of edge cases, insurance billing quirks, scheduling rules, regulatory forms. An agent doesn't magically know those; it has to be taught, and the teaching material is the incumbent's proprietary asset. This is why vertical players have a better defense than horizontal ones, a theme worth examining on its own.

So the honest answer is: agents commoditize the experience of software faster than they commoditize the substance of it. The buttons get cheap. The truth underneath them does not.

The Value Migration: From Screens to Outcomes

Here's the shift that actually matters, and it's subtler than "SaaS dies." Value is migrating up the stack, from screens to outcomes, and whoever captures the outcome layer captures the pricing power that used to live in the interface.

Think about what a buyer is really purchasing. Nobody wanted a CRM; they wanted closed deals and clean pipeline data. Nobody wanted a ticketing system; they wanted resolved customer problems. The software was a means. For years it was the only practical means, so it got to charge as if it were the end. Agents break that monopoly on means. If an agent can deliver the resolved ticket directly, the buyer will increasingly pay for the resolution, not the seat.

This is why the conversation keeps circling back to the disappearing dashboard. A dashboard exists to inform a human who will then act. Compress the inform-and-act loop into an agent and the dashboard becomes vestigial, interesting for audit, useless for operations. The application layer doesn't vanish; it goes invisible. It becomes a substrate the agent operates against rather than a surface a human stares at.

The strategic implication for vendors is brutal in its clarity: if your differentiation lives in the interface tier, you are differentiating on the thing being commoditized. If it lives in the data and outcome tiers, you're sitting on the thing capturing the new value.

Pricing Power Is the Real Battleground

Commoditization shows up in the income statement before it shows up anywhere else, and the leading indicator is seat count.

The seat-based model assumes a human operates the software, and you charge per human. Agents quietly detonate that assumption. When one agent does the work of a five-person team, the buyer doesn't need five seats, and a vendor charging per seat just watched 80% of an account's revenue evaporate, even if the underlying software is used more than before. McKinsey's research on the economic potential of generative AI frames this as a labor-productivity shift, and that framing is exactly the problem for seat-priced vendors: the value moves from number of users to amount of work done, and those two numbers are decoupling fast.

This is the core of why "agents commoditize the application layer" is really a pricing claim wearing a technology costume. The software might be just as essential as before. But if it can't capture value when humans stop being the unit of consumption, its economics get commoditized even if its function doesn't. That's the trap. Repricing toward consumption, outcomes, or a hybrid is not optional for exposed vendors; it's survival. The ones moving early, repricing for the agent era deliberately rather than defensively, are trying to get ahead of the seat-count shrinkage their investors have already started to fear.

Who Wins, Who Gets Squeezed

Sorting the field:

Squeezed hardest: horizontal workflow tools, RPA, generic no-code, integration middleware, and any "thin wrapper" whose entire value was a nicer front end over an API someone else owns. If a competent engineer would call your product "a UI over a database with some rules," you're in the commoditization zone.

Pressured but defensible: horizontal systems of record (CRM, ERP, ticketing) that own real data. They lose interface and seat revenue but keep the data moat, and many will reprice around their own agents rather than be disrupted by outsiders. The risk is they cannibalize seats slowly while a startup does it fast.

Best positioned: vertical SaaS with deep proprietary data and regulatory entanglement; systems of record with governed write-access; and the emerging agent platforms that own the outcome and the customer relationship. The new prize is becoming the "agent of record", the trusted layer a customer routes work through, which carries its own lock-in.

The uncomfortable truth for incumbents is that distribution, not capability, is their real advantage. They already sit inside the enterprise, hold the data, and have signed contracts. A startup with a marginally better agent still has to displace all that. Capability is becoming cheap; distribution and data access are not.

A Practical Framework for Founders and Buyers

If you build software, ask three questions. Where does my differentiation live, interface, workflow, logic, or data? If it's in the top two tiers, assume commoditization and move your value down. Can I capture value when the consumer is an agent, not a human? If your pricing dies the moment seats shrink, reprice now. Do I own data or authority the agent depends on? If yes, you have a moat to build the agent into, rather than a moat to be disrupted out of.

If you buy software, the mirror questions apply. Am I paying for an interface I'll soon bypass? Is this vendor repricing honestly for the agent era, or bolting on an "AI" badge and hoping I won't notice? The difference between genuine transformation and lipstick is whether the vendor will let an agent, including a competitor's agent, operate against their data on fair terms. The ones gatekeeping integrations are telling you exactly how worried they are.

The application layer isn't dying. It's being unbundled into the parts that were always commodity (the screens) and the parts that never were (the truth and the right to act). Agents are just the force finally separating them.

Insights Most People Overlook

1. The most "AI-proof" software is often the most boring. Everyone fixates on flashy, intelligent-seeming tools as the survivors. The opposite is true. The dull system that holds the regulated ledger, the audit trail, and the signed compliance attestation is nearly impossible to commoditize, precisely because its value was never the experience. Boring infrastructure inherits the earth; clever interfaces get absorbed.

2. Commoditizing the interface can raise the value of the data underneath. When agents make the front end disposable, every vendor's data becomes the contested asset, which means owning clean, well-structured, permissioned data is worth more in the agent era, not less. The interface losing value and the data gaining value are the same event viewed from two sides.

3. The dangerous position isn't being disrupted, it's being "used." A vendor can keep all its customers, see usage go up as agents hammer its APIs, and still lose because it priced per seat and the seats vanished. Healthy usage metrics can mask a collapsing business model. Watch revenue-per-account against API-call volume; a widening gap is the early warning.

4. Incumbents may commoditize themselves faster than agents would. The "incumbent's dilemma" cuts deeper than usual here. To compete, incumbents ship their own agents that bypass their own seat-based pricing, actively dismantling their highest-margin revenue to avoid losing it to a startup. The disruption is partly self-inflicted, and it's rational.

5. "Outcome pricing" is harder than the slide decks admit. Selling outcomes instead of software sounds clean until you have to define, measure, and attribute the outcome when an agent, a human, and three other tools all touched it. The vendors that win the repricing war won't be the ones with the boldest outcome promise; they'll be the ones with the most defensible measurement of outcomes. Attribution is the unsung moat of the agent economy.

References

• The Economic Case for Generative AI and Foundation Models, Andreessen Horowitz
• What's Ahead for Software Engineering With GenAI, Gartner
• The Economic Potential of Generative AI: The Next Productivity Frontier, McKinsey