SaaS is evolving, not dying

There has been a lot of talk during the last two years about the Agentic era leading to the death of Software as a Service aka SaaS. In June 2024, the Salesforce stock slid more than 20% based on its quarterly earnings. It became especially loud recently when Satya Nadella recently stated his views in this interview

The argument is that all SaaS business at its core is a database and an application wrapped around it. The application contains business logic and dashboarding capabilities. A well trained agent based on an LLM should be able to replace the business logic in the future.

Indeed, pure SaaS platforms, where the applications are the classic CRUD (Create, Read, Update, Delete) actions on a database and some visualization built on top – are becoming commoditized. These platforms can be built in google cloud tools in a matter of days.

In theory, one could build an agent, connect it to an existing (set of) SaaS application(s) and use it, and thereby kill the app interface. There are following challenges 

Missing Domain expertise: SaaS applications, possess not just data, but also domain expertise. This expertise is built into the complex workflows and business processes. For e.g. there are workflows that determine how a new lead is qualified in a CRM application. In a real automation, the agent will automate the existing workflows everytime a new lead needs to be qualified.

Relying on user prompts: People imagine that an Agent can be provided connectivity to a tool and thereby kill the SaaS interface. However, the product experience would still rely on the user’s prompting skills. We know very well the challenges of asking imprecise or very complex questions to an over eager intern. They will rely on their general knowledge and desire to please you to provide you with great – but unhelpful responses. In some cases one might still require the user to provide login credentials, and then the prompts. 

MCP:

I think the more likely approach will be to have agents as native experience within the SaaS app itself. Using MCP protocols, we will build architectures that call on SaaS Agents to automate tasks within that platform for us. 

SaaS is a system of engagement built on a system of records. At least in the near future, agents will not replace the system of engagements. It is likely that SaaS agents will offer an improved engagement and automation.