Alexey Spas is the founder and CEO of Instinctools, a software engineering company focused on AI-powered digital solutions.gettyBefore agentic AI came into the picture and caused a dramatic turn of the tables in the tech scene, there was point-and-click software-as-a-service (Saas). Once an investor darling and a cash-flow king, SaaS tools have traditionally operated as sleek wrappers around a database, and the user’s role boiled down to manually clicking through the interface to get a task across the finish line. ​With AI agents stepping up as autonomous researchers, the relevance of the go-between dashboard has been called into question with dozens of apocalyptic headlines warning about how SaaS has been rendered obsolete. However, this definitive prognosis ignores the fact that incumbent SaaS tools are more than just a user interface (UI). They are essentially systems of records with their own business logic, permission structures and deeply entrenched enterprise workflows, which will not easily be dislodged by AI.​​From this perspective, I see the future of SaaS playing out according to one of these three scenarios, though each comes with its own challenges.1. SaaS tools are vibe-coded out of business.The first scenario is based on the clean and catchy premise that software becomes infinitely commoditized because of AI agents. In this case, with enterprises using AI to DIY versions of tailored tools at a near-zero code cost, the carefully constructed moats of subscription-based software companies will inevitably be priced toward obsolescence.​While this narrative plays well in headlines, its underlying rationale rests on a conceptually flawed understanding of what initially made SaaS a centerpiece in the enterprise tech stacks. Companies aren’t paying vendors just for syntax. They are licensing distribution, reliability, mature infrastructure, global support systems, hardware and economies of scale across millions of customers. ​Even if AI slashes the marginal cost of building enterprise software, models won’t take care of the operational overhead required to run and maintain it. Companies would have to support, update, secure and distribute homegrown tools in-house—costs that SaaS vendors already amortize across vast customer bases. In this case, the unit economics don’t hold up.​Another clincher that makes a wholesale migration from SaaS to vibe-coded tools a stretch is the fact that enterprises store critical business data inside vendor platforms. Not only do the model makers not offer databases to re-host that data, but a rip-and-replace migration of this layer would also be costly and operationally taxing.​2. SaaS systems hand over the keys to outside AI agents.The second scenario paints a more evolutionary future for vendors. According to this narrative, existing systems of record remain intact, but the interface layer is getting externalized to AI agents that become part of companies’ multi-agent systems. Instead of combing through UI layers themselves, users assign internal agents to interact with SaaS platforms and retrieve data for them.​From a technical standpoint, this course of events pushes SaaS architectures down the API-first and event-driven design path. Create-read-update-delete operations would move into the background as workflow orchestration APIs take priority, while webhooks, message queues and real-time state synchronization would become the main integration surfaces. Taken to its logical extreme, the model decouples software usage from human interfaces altogether and turns vendors into agent endpoints.​While the second scenario is more plausible than the first one, it overlooks an important prerequisite: Agents can become only as autonomous as SaaS systems will allow them to. In practice, lifting the gates to outside agents upends the existing model of human-centered authentication and expands the attack surface where API keys and OAuth tokens can be easily exposed at scale.​Some incumbents have already made moves that put a question mark next to this scenario. For example, SAP’s new API policy has walled off platform data to everyone but SAP-endorsed access pathways. Meanwhile, ServiceNow introduced a metered layer that outside AI agents must pass through to operate on platform data and execute workflows. Until this division exists, this scenario won’t play out to the fullest.3. SaaS Becomes services-as-software.In this final scenario, incumbent SaaS vendors embed AI agent capabilities directly into their products as a defensibility strategy. For platforms, this is a point-of-leverage master plan because they already own the environment where work happens. So, vendors are able to lay the boundaries of governed execution, not just the interface for it. ​Once that boundary is owned, the SaaS system of record structurally segues into a services-as-software system for work. To this end, AI agents become just a new execution layer as SaaS platforms internalize agentic interfaces within the existing system of record.​Many heavyweight vendors like Salesforce and HubSpot have already jumped on this bandwagon by debuting their very own forms of conversational support. Absorbed into the product surface, the agent layer brings together the data plane and the action plane, transforming—but not deprecating—the moat of these SaaS tools.​Potentially, this scenario could put a strain on the current “per-seat” pricing model used by majority of SaaS tools. Agentic systems incur inference, computation and orchestration costs, which would necessitate a new, pay-per-transaction pricing model from vendors. That open-ended consumption shift, however, wouldn’t catch on easily with companies because of a higher and less predictable bill. That’s why, for now, agent economics is kept under wraps by vendors.ConclusionAs headlines herald the coming “SaaSpocalypse,” the reality is that AI agents will likely keep the SaaS market on its feet, not throw it to the curb. From my work with enterprises, the future shapes up to be more hybrid than black-and-white. There are still regulatory-heavy sectors that shun fully autonomous AI agents because of regulatory requirements. Other, less regulated sectors double down on building out their own multi-agent systems. The headline is: What we’re witnessing now is less of an extinction event for SaaS and more of an evolutionary filter.​Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?