Will Agents Disrupt SaaS as SaaS Disrupted On-Prem?

Over the past decade and a half, web APIs have defined how software connects, scales, and compounds. They abstracted complexity and created a common language for systems to talk to one another. The rise of API-first companies such as Twilio, Stripe, and Plaid showed how modular interfaces could turn infrastructure into reusable products. Developers could “call capabilities” as easily as they once imported libraries. This unbundling of monolithic software into smaller, interoperable parts became the foundation for modern SaaS and, ultimately, the programmable economy.

That same factorisation principle was at the heart of my earlier writing on The Lazy Programmer and The Shipping Container Moment for Software. Both explored how software engineers and entrepreneurs create leverage by simplifying what used to be complex, fragmented processes into clean, reusable units. The shipping container turned global trade into an abstraction layer—standardised, portable, and scalable. APIs did the same for software. And now, the Agents are unlocking capabilities for AI natives.

New protocols such as MCP and A2A represents the end of the request-response era and the beginning of a conversational, context-driven one. APIs modularised logic; MCP modularises tasks. The shift feels familiar: we once designed endpoints, now we design agents that can interpret and negotiate goals dynamically. Instead of a one-off transaction, software is becoming an ongoing conversation between systems.

The SaaS revolution was never just about moving software off-premise. It succeeded because it changed the economics of distribution, allowed new incumbets to redfine entire categories of software. SaaS enabled vertical solutions—tailored systems for industries such as finance, logistics, and healthcare—to scale globally at near-zero marginal cost, without armies of consultants deploying on-site. The real disruption came from packaging expertise into scalable delivery. In the same way pehaps, agent-native startups will challenge today’s SaaS incumbents. They won’t win because they’re “non-SaaS”; they’ll win because they deploy reasoning vertically, embedding intelligence directly into context. Some SaaS players will adapt quickly. Others will move too slowly, bound by their old abstractions.

To understand the opportunity, consider a simple example. Imagine I’m sitting in my car, speaking to my preferred LLM, and I decide I want to book a meeting room. In the current world, I’d have to open an app, sign in, and navigate through a SaaS product designed for co-working space management. In an MCP world, my model would simply query a “Meeting Room MCP Server” that exposes availability, pricing, and booking capabilities in a structured, context-aware format. I could build that server once and sell it to every co-working space on the planet. That’s not a feature—it’s an entirely new business model. There’s a startup idea right there: a universal meeting-room protocol that plugs directly into any conversational agent.

Of course, there are risks. Scaling MCP securely will be a major challenge. Many of today’s identity and access frameworks were built for static, human-initiated interactions, not for autonomous agents negotiating across systems in real time. The question of how to manage personal identifiable information in an agentic world remains unresolved. How does consent work when the actor is not a human, but a model acting on one’s behalf? How do we enforce data boundaries when context, not code, is the integration layer? These are open problems, but many companies are already at the forefront of solving them—among them, Prefactor and FrankieOne, who are pioneering new approaches to contextual identity, verification, and privacy-aware orchestration. Their work will be critical to ensuring MCP evolves within a secure, compliant framework.

The implications are profound. Just as APIs enabled horizontal scale by exposing functionality, MCP enables cognitive scale by exposing understanding. APIs built network effects through data and distribution; MCP builds them through shared context and reasoning. The next generation of winners will not simply run on APIs—they’ll think through MCP.