The SaaS Layer Is Dissolving

tl;dr: The application layer of enterprise software, all those SaaS products you pay for monthly, is being restructured by agentic AI. This is already happening. Technical teams and startups are building this way now. The strategic question is whether you're positioned as a winner or a casualty when the rest of the market catches up.

The cost of building bespoke internal tools is falling below the fully burdened cost of SaaS subscriptions
Infrastructure and APIs survive; generic workflow products don't
Your moat is your data, your integrations, and how fast you update your model of where software value lives
The window to reposition is two years, maybe less. Companies that don't move on this may not get another chance

The Shift

For thirty years, enterprise software has followed a stable architecture: infrastructure at the bottom, applications in the middle, users at the top. The application layer has been where the money lives. Salesforce, Workday, ServiceNow, and HubSpot are billion-dollar companies built on the premise that building software is hard enough to justify renting it in perpetuity.

That premise is losing its foundation.

Agentic AI tools don't just write software faster. They generate complete working applications on demand, tailored to exact requirements. Non-technical operators are building functional internal tools in hours. These aren't polished products ready for commercial release, and they don't need to be. They solve the specific problem, they cost a fraction of what a SaaS license costs over time, and they can be modified or discarded as requirements evolve.

We're seeing this across our client engagements: teams that used to evaluate vendors are now evaluating whether they need vendors at all for entire categories of tooling.

What Survives, What Doesn't

The application layer won't vanish entirely, but the value distribution within it is about to shift dramatically.

What persists: Infrastructure. ERP systems, databases, authentication services, payment processing, and anything else requiring deterministic behavior, transactional integrity, or functioning as a system of record. These are foundations, and they remain necessary.

What grows: APIs as the universal interface. The collapse of the application layer doesn't mean the collapse of capabilities. It means those capabilities get exposed as composable primitives that agentic systems assemble on the fly. Stripe becomes a transaction primitive that an AI calls when needed. The businesses that thrive will be the ones that make themselves composable.

What dissolves: Generic workflow products. Visualization tools, marketing automation platforms, project management software, and the whole middle tier of nice UIs layered over standard processes: this is where the displacement hits hardest. When an AI can recreate that functionality in minutes, configured exactly to your process rather than forcing your process into someone else's product decisions, the value proposition of $20/seat/month for features you mostly don't use collapses.

The Strategic Problem

Most organizations are still running on a mental model of software value that was formed in the early SaaS era. That model says: software is hard to build, so you buy it; you accept the vendor's product decisions as a reasonable trade-off for not having to build and maintain your own; and your technology strategy is largely a procurement exercise.

Every piece of that model is breaking simultaneously.

The organizations that recognize this early gain asymmetric advantage, largely because they've updated their understanding of where value lives in the stack before everyone else does. The ones that cling to the old model are accumulating strategic debt while the cost of switching rises.

This is a pattern we see repeatedly: the highest-leverage move is updating the organizational model of how the landscape actually works. Everything downstream, from vendor strategy to build-vs-buy decisions to hiring to competitive positioning, follows from that update.

Implications by Position

If you sell SaaS: Your runway is shorter than your roadmap assumes. The strategic play is to become infrastructure (boring but durable) or become so embedded in customer data that you're irreplaceable. Anything in between, the nice interfaces and generic workflows and seat-based pricing, is increasingly exposed. The question to pressure-test: if an AI could compose any functionality on demand, what would it still need from you that can't be generated?

If you buy SaaS: Your competitive advantage lives in your data and what you can do with it. "Nobody ever got fired for buying Microsoft" was safe harbor for a long time. That's eroding. The investment priorities are data infrastructure, API literacy, and building organizational capacity for AI-native workflows.

If you're building something new: Stop thinking about applications as products. Think about capabilities as services. The businesses being built today that will matter in five years are the ones that make themselves useful to agentic systems: composable, callable, embedded in workflows rather than sitting behind a login screen waiting for a human to click through.

The Window

The full dissolution will play out over five years, roughly how long enterprise procurement cycles take to catch up with technological reality. But the strategic window is two years, maybe less. Early adopters are already building this way. By the time the broader market catches up, the organizations that moved early will have compounding advantages in cost structure, speed, and adaptability that latecomers can't close.

Enterprise procurement cycles are five-year animals. If you need two of those years just to recognize the shift and begin planning, you're already behind. The organizations that survive this transition are the ones defining their strategy now and beginning to move, while the cost of waiting compounds silently.

The commercial application layer was always an artifact of a specific technological moment: when building software required deep specialized skills and enormous time investment. That moment is ending. What replaces it is a thinner coordination layer between human intent and machine capability, assembled for each problem, adapted as requirements shift, free from someone else's product roadmap.

This is happening. The only variable is where you are when it arrives.