Is SaaS dead in the age of AI?
What types of SaaS applications are most likely to survive?
The following is an excerpt from the Q2 2025 Investor Letter
The tech newscycle is abuzz with how GenAI-driven code generation will allow for customized AI agents or bespoke AI apps to be written for every enterprise’s needs, and replace the SaaS applications. Satya Nadella famously made the comment that SaaS is dead. The arguments for Agentic AI are compelling: AI Agents address many of SaaS’s core limitations by eliminating data silos and automating workflows end-to-end across enterprise workflows. They also make software intelligent, autonomous, and outcome-driven rather than merely assisting humans.
AI Agents are smart but will need supervision and controls
An application serves as a container for business logic, and SaaS apps - or their predecessor, self-hosted client-server applications – enforce boundaries on what actions can be performed for business workflows. AI agents or AI apps themselves will need similar boundaries governing what they can access or act upon for compliance and security reasons. Such boundaries require application-like constructs and may lead to the emergence of new AI applications, though a few SaaS innovative applications may also rise to meet this challenge.
The data silos problem is organizational, and not due to proliferation of SaaS apps.
While people complain the proliferation of SaaS applications has created data silos, the problem is often not technology-driven. Different departments within a company - from legal to marketing - want control over their datasets because they’re solving distinct problems. The truth is that SaaS apps didn’t create the data silos problem. Achieving a single source of truth for data has been a challenge for enterprises, from enterprise data warehouses to data lakes.
Complex business workflows, vertical SaaS will be harder to replace than simpler ones
AI agents may initially replace SaaS apps for project or calendar management, as these business workflows are standardized across industries and don’t face significant compliance or legal hurdles. In contrast, workflows for payroll or specialized industries like manufacturing will take longer to be replaced by AI. Vertical SaaS apps have the advantage of well-organized datasets for complex workflows. These datasets are internal to enterprises and unavailable for most Large Language Models (LLMs) to train on, giving vertical SaaS an edge in the race to become AI-ready.
Key Takeaway: Future may be hybrid - SaaS Apps, and AI Apps, Agents likely to co-exist
Tilden finds it more useful to consider probabilities and looks to history for patterns that may repeat. History shows us that trends evolve gradually rather than abruptly: SaaS apps have not replaced all client-server apps, and cloud computing hasn’t eliminated self-hosted solutions. Similarly, SaaS apps may coexist with bespoke AI apps and AI agents for the near future.
