How AI Agents Will Disrupt SaaS in 2025 | by Ori Ziv | Jan, 2025 | Medium

How AI Agents Will Disrupt SaaS in 2025

Ori Ziv

9 min read

1 day ago

Created with DALLE-3

The world of Software-as-a-Service (SaaS) is on the brink of a seismic shift. With the rapid advancements in artificial intelligence, and especially LLMs (Large Language models) AI agents are emerging as transformative tools poised to redefine how SaaS platforms operate, deliver value, and interact with users. By 2025, we can expect AI agents to not only enhance existing systems but also drive the creation of entirely new business models.

Recently, the concept of Vertical AI Agents has gained significant traction, suggesting a potential paradigm shift even more transformative than traditional SaaS. As highlighted in the Lightcone podcast by YC, these specialized agents, designed to deeply integrate with specific industries and use cases, are poised to create an entirely new category of business opportunities. With projections hinting at the emergence of hundreds of billion-dollar companies in this space, Vertical AI Agents could outscale SaaS by an order of magnitude

Satya Nadella, CEO of Microsoft, has offered a visionary perspective on how AI agents will reshape the SaaS landscape, predicting a fundamental shift in the way business applications are conceived and utilized. In a recent B2G podcast interview, he suggested that the traditional structure of SaaS — essentially CRUD (create, read, update, delete) databases governed by business logic — could collapse in the era of agentic AI.

In this blog post, I’ll try to dive into the world of AI agents and explore:

What AI agents are?
The evolution from single-agent systems to multi-agent AI systems.
Platforms enabling the design and deployment of AI agents.
How human-agent interfaces will evolve?
Why AI Agents has the potential to disrupt the SaaS market?

What Are AI Agents?

AI agents are software programs powered by artificial intelligence ,usually based on a foundation model, designed with a purpose to perform specific tasks autonomously. These agents are context-aware, goal-oriented, and capable of learning from interactions with their environment or users.

Think of AI agents as virtual assistants, but far more specialized and capable. They can automate repetitive tasks, analyze complex data and make decisions, and interact with users and systems to achieve defined objectives. For becoming autonomous, they will be equipped with tools and capabilities such as function calling, memory, ability to search the internet, or performing CRUD operations at external systems.

For instance, an AI agent in a SaaS CRM platform could independently manage customer interactions, schedule follow-ups, and provide predictive analytics for sales teams.

Anatomy of AI Agent

At its core, an AI agent is built upon a foundation model, typically one of the large language models (LLMs) such as GPT4, Claude, Gemini etc or the small and low cost language models (SLMs) that most suitable for the specific task the agent was designed for. The system prompt acts as the guiding framework, defining the agent’s purpose and the format of its outputs. To make the agent contextually relevant, it is connected to an external knowledge base or data source, grounding its responses with accurate, domain-specific information. A common approach for this integration is the Retrieval-Augmented Generation (RAG) pattern, which combines retrieval of external data with generative capabilities. Beyond its foundational understanding, the agent is equipped with a toolkit — specialized capabilities and skills enabling it to autonomously perform actions, trigger workflows, or solve tasks aligned with its objectives. Coordinating all these components is the orchestrator, the glue that binds the agent’s functionality. The orchestrator processes inputs from users, manages internal operations, and delivers coherent results either directly to the user or to other agents in systems involving multi-agent interactions.

At last, an optional user experience will expose the agent capabilities to the end user, if a user interaction needed. In multi agent systems some agents will only communicate internally with other agents in the system, while others will communicate with the end users.

Agent anatomy (source: Microsoft Ignite 2024)

From Single-Agent to Multi-Agent AI Systems

In 2025, we’ll see a significant transition from single-agent AI solutions to multi-agent systems.

Single-Agent Systems: These are focused, task-specific AI models, like smart chatbots. While effective in isolated scenarios, they have limitations in handling complex, interconnected workflows. Single agent systems usually require human in the loop, that provides continues feedback.
Multi-Agent Systems: These involve a network of AI agents collaborating to solve problems or achieve goals that require diverse expertise. Think of a team collaboration that internally communicates with each other, criticize each other, and improve each other results in order to solve a given task, vs a single agent that only get’s its feedback from the human that communicates with it.

For example, in a project management SaaS platform, one AI agent might prioritize tasks, another could forecast project risks, and a third might handle resource allocation — all working in harmony. These systems emulate human teams, where each agent brings a unique specialization.

Multi-agent systems are particularly compelling because they can dynamically adjust to new challenges, delegate responsibilities, and even negotiate with each other to optimize outcomes.

Multi-agent system is aiming to reduce human in the loop to the minimum, asking the human supervision and/or approval with taking actions only when needed or designed to.

Platforms and tools for Building and Designing AI Agents

The proliferation of AI agents has already fueled by platforms that make their design, training, and deployment accessible. These platforms typically provide:

1. Pre-Built AI Models: Ready-to-use pre trained models and ready to be deployed in the cloud.

2. Customizability: Tools for fine-tuning AI agents to fit unique business needs.

3. Evaluation: Tools for evaluating the quality and safety of the AI agent, and its communication within the environment.

4. Integrations: APIs and connectors to integrate AI agents into existing SaaS platforms seamlessly.

5. Multi-Agent Frameworks: Advanced platforms offering templates and protocols for multi-agent collaboration. These frameworks often include the ability to simulate the behaviour of multi agents within a system.

Some emerging platforms and tools to watch include:

Semantic Kernel: A lightweight, open-source development kit by Microsoft that lets you easily build AI agents and integrate the latest AI models into your C#, Python, or Java codebase. It serves as an efficient middleware that enables rapid delivery of enterprise-grade solutions.

LangChain: Like semantic kernel, LangChain is an OSS toolkit for building applications powered by language models, often used for designing conversational agents. They offer Javascript and Python SDKs for programming.

OpenAI’s Assistant APIs: With GPT-4o, o1 and beyond, developers can build increasingly sophisticated agents and deploy them in their applications or in the ChatGPT platform.

OpenAI’s Swarm: An educational framework developed by Open AI team exploring ergonomic, lightweight multi-agent orchestration for research and simulation purposes.

AutoGen: A multi-agent conversation framework by Microsoft Research that simplifies building LLM workflows, enabling diverse applications across various domains. It provides optimized LLM inference APIs to enhance performance and reduce costs, making it a powerful tool for creating efficient and versatile AI-driven solutions. AutoGen is also offering “AutoGen Studio” for no-code/low code experience of constructing multi agents.

Vercel’s AI SDK: A free open-source library by Vercel that gives you the tools you need to build AI-powered products with Typescript and Node.JS that is easily integrated into Next.JS web apps.

AutoGPT: Open-source framework for autonomous, multi-agent systems development with CLI.

TinyTroupe: Interesting OSS project by Microsoft, provides a language models-powered multi-agent persona simulation for imagination enhancement and business insights.

No code/Low code SaaS-AI BuildersSpecialized tools that allow non-technical users to build, deploy and configure AI agents.

Azure AI Foundry is a robust platform for building, deploying, and managing AI-driven applications using a variety of models and services. it offers a new assistant builder (currently in public preview) for building and deploying agents in Azure.

Microsoft Copilot Studio allows you to create and customize AI assistants for Microsoft 365 and other channels using low-code tools and generative AI.

These platforms are democratizing AI development, enabling businesses of all sizes to harness the power of AI agents.

How human-agent interfaces will evolve?

For AI agents to succeed, they must seamlessly integrate into workflows and be user-friendly. This is where human-agent interfaces (HAIs) come into play.

Conversational Interfaces

Conversational AI will redefine the way users interact with SaaS platforms. Instead of manually navigating through menus or forms, users can simply type or speak requests, such as “Create a report for last month’s sales,” or “What’s the status of project X?” The AI agent will understand the query, interpret intent, and execute commands. This reduces friction, enabling users to get things done faster and with much less cognitive effort. In addition, these conversational agents will become more context-aware, maintaining ongoing dialogue across multiple interactions, which can improve user experience.

Proactive Agents: Instead of waiting for input, AI agents will anticipate needs and initiate actions, such as suggesting optimizations or alerting teams to anomalies.

Personalized interface: AI agents will tailor SaaS interfaces to individual users by learning from their behavior, preferences, and needs. Instead of a one-size-fits-all approach, the platform will adapt dynamically to each user. For example, a SaaS platform for project management might present different dashboard layouts to different team members based on their roles — managers could see a high-level overview, while developers might get more detailed tasks and timelines. Over time, AI will suggest custom shortcuts, tools, or workflows that match each user’s patterns, making the interface feel personal and intuitive.

Augmented Reality (AR)

Augmented reality could significantly enhance SaaS interfaces, particularly for tools focused on design, visualization, and collaboration. Imagine using a SaaS application for data visualization where an AI agent overlays 3D models or charts in the user’s physical space, or using AR to project a virtual workspace for team collaboration. For instance, in a SaaS application for architectural design, users could view and manipulate building plans in real-time, visualizing how changes would look in their environment. AR combined with AI could create an immersive experience where users can interact with complex data or designs in more meaningful and tangible ways. A well-designed HAI bridges the gap between sophisticated AI capabilities and end-user accessibility, ensuring AI agents complement human expertise rather than replace it.

Decentralized InterfacesRather than having users interact exclusively within a single app, the AI agent could become a bridge between different ecosystems. For instance, an AI agent might allow users to seamlessly access SaaS functionality from their messaging platform or virtual workspace without needing to open the dedicated app. Imagine having a conversation in Microsoft Teams, and the AI agent assists by fetching customer data or generating reports without ever leaving the platform. This decentralization could improve accessibility, as users wouldn’t be tied to a single platform or app to complete their tasks.

UX Layout shift

The user experience and layout of SaaS platforms will undergo significant transformation to accommodate the integration of conversational interfaces. Traditional UI elements will give way to more fluid, personalized designs, seamlessly embedded within conversational canvases. This shift will streamline interactions, reducing the reliance on static menus and forms, and creating a more dynamic, intuitive experience tailored to individual user needs.

Human-Agent communication

When designing Human-Agent Interfaces, several factors must be considered to ensure effective and seamless interaction. The agent’s goals and user preferences should be clearly defined and respected, with mechanisms for users to provide feedback for improvement. To enhance user understanding, the interface should help users verify the agent’s actions, convey consistent behavior, and tailor the level of detail based on context. Past interactions should inform the agent’s communication to build trust and continuity. Communication should address the agent’s current actions, future intentions, and outcomes, including whether the goal was achieved and any side effects. Designing such interfaces also requires tackling challenges in both conveying information to users and interpreting instructions from them to ensure alignment and a positive user experience.

source: Challenges in Human-Agent Communication, Microsoft Research

The Disruption Ahead

In 2025, AI agents might potentially disrupt SaaS in profound ways:

1. Operational Efficiency: Automating repetitive tasks and providing real-time insights will free human agents to focus on strategic work.

2. Personalization at Scale: SaaS platforms will deliver hyper-personalized experiences for every user, thanks to AI agents’ ability to learn and adapt.

3. New Business Models: AI agents as a service (AIaaS) will emerge, allowing businesses to lease agents specialized for specific tasks.

4. Competitive Differentiation: Companies that integrate multi-agent systems and advanced HAIs will gain a significant edge in the SaaS market.

Conclusion

AI agents are not just tools — they are collaborators, amplifiers, and disruptors, embodying a new era of “superagency” in how we work and innovate. As we move toward 2025, businesses that embrace this transformation will thrive, while those clinging to traditional SaaS paradigms may struggle to stay relevant. The question is not whether AI agents and multi agent AI systems will disrupt SaaS — it’s how prepared organizations will be for this revolution.