The term SaaS-pocalypse has been popularized lately to summarize the idea that the traditional business software providers’ business model is about to undergo a tectonic structural shift because artificial intelligence will change how work is performed, rendering applications and their subscription pricing models obsolete. In this telling, AI assistants or agents can reduce the need for humans to spend time inside multiple applications. Instead of people clicking through multiple tools, agents will execute tasks across systems, using application programming interfaces (APIs) and process automation. The result would be a reduced number of seats necessary for a given enterprise because a defined orchestrated value chain will obviate the need for standalone applications. Consequently, application software revenues will wither as agents and AI displace their value-add.
I disagree. I see some of the thinking behind this rooted in the new-paradigm-sweeps-incumbents-aside model that had relevance in the 20th century but not today. So, a basic
Another underlying assumption here is that business software providers will be incapable of or unwilling to change their packaging and pricing models. The reality is nobody has a handle on pricing at the moment. The market will dictate how this evolves over the next couple years, and software providers will need to be flexible in adjusting to competitive pressures. Since we’re at the start of the S-curve in agent adoption, the impact of pricing on the bottom line over the near term therefore will be muted.
As the 18th century essayist, Samual Johnson observed (and Warren Buffett quoted), a horse that can count to ten is a remarkable horse, not a remarkable mathematician. In other words, don’t confuse an astonishing and unexpected ability with true expertise. Writing contracts well-versed in the law and consistent with an enterprise's requirements is exactly what a large language model (LLM) is designed to do. LLMs can also be the red team that looks for vulnerabilities in contracts and maybe even intentionally inserts clauses and wording that disfavor the other party. LLMs are great at democratizing coding. They can do all sorts of creative things, but they can’t do everything.
Similarly, service providers are well equipped to create and manage agents that orchestrate cross functional end-to-end value chains as a unified workflow, but that doesn’t mean they are also the best choice for agents that will execute specific tasks along the way. This will be especially true if a business application already has its own agents for the tasks or sub-tasks embedded in their software. You can’t replace core business application functionality with ad hoc agentic AI. For example, in an order-to-cash process, the business application is better positioned to check material, capacity and scheduling availability and crucially, recommend workarounds as issues arise. Or analyze product characteristics to correctly classify goods and recommend tariff numbers for international shipments. “Close enough” is fine in a game of horseshoes, but not in this case, or in accounting or supply chain management. “Close enough” doesn’t address governance and quality issues. Moreover, which actor is best positioned to ensure that rules used for optimizing inventory reflect a company’s current strategy or that the tariff codes are up to date? Spoiler alert—it’s the application with the domain expertise and the focus on ensuring that models and agents are performative and the data is timely and reliable.
Another assumption underlying the impending doom of the SaaS software industry is the ability of AI to simplify writing code. This is especially the case for enterprises that have legacy code and need to address technical debt issues. In any large enterprise there are likely to be multiple edge cases where a bespoke application is the right choice, but this isn't the 1980s. The reason back then for building your own application was to achieve competitive advantage. Today, however, the economic rationale for developing bespoke software is to avoid being at a competitive disadvantage because none of the many off-the-shelf applications meets the real operational needs of the enterprise.
Service providers or IT departments can do a great job building agents that orchestrate cross-application workflows, but these will rely on existing SaaS applications to execute the specific elements of a process. There's plenty of scope for those players to add value with agents and AI, but they cannot economically replicate the specific deep functionality of most of the significant business software applications. And you have to wonder why you would want to develop a bespoke agentic system to handle, say, your travel and expense requirements if you can get it off the shelf for the same or less without having to support and maintain it.
That the assumptions underlying the meme are already verifiably false doesn’t matter to the tellers of the SaaS-pocalypse. It’s a great attention grabber built on long obsolete 20th-century paradigms, and it’s plausible only if you have been paying no attention to what’s happening in the business software world. And then only if you trivialize the decades of domain-specific development that’s gone into these applications and ignore how work is actually done and by whom in an enterprise.
I strongly recommend that service providers and IT departments thoroughly acquaint themselves with what’s available in software applications before investing in agent-led systems to handle end-to-end value chains. There is plenty of scope for orchestrating end-to-end value-chain automation systems, but this mainly will act as the connective tissue that ties together the business applications that execute the work. I also recommend that business application providers do a better job of highlighting the capabilities and future roadmaps.
Regards,
Robert Kugel