Recent advancements in Generative Artificial Intelligence (AI) combined with the rise of APIs and microservices have created an opportunity to transform traditional business models and technical systems. Composable solutions can play a critical role in strategic solutions where agility, scalability, modularity, and interchangeability are crucial. In this context, for most enterprises, AI shouldn’t be viewed as a monolithic solution (i.e., the solution is an AI).

Instead, the solution is made up of components, including the ERP, CRM, specialised calculators, and specialised AI agents that address discrete tasks. Components can vary in size and scope to provide reusable capabilities, services, and even micro-services.

Some readers might look askance at the reference to monolithic applications like ERP and CRM listed above. Typically, ERP and CRM solutions provide mature software solutions that are configured to purpose. They do not typically provide the “secret sauce” that helps your business 10x its revenue or cost savings. Because of this, most businesses should be able to use off-the-shelf software for these applications.

By building composable solutions, businesses can create adaptable, intelligent systems that better cater to specific requirements while seamlessly integrating the components into a broader platform or solution.

I recently met with the board member of a Fintech that specialises in estimating taxes for high-net-worth clients for every country in the world where they have holdings. This very specialised and complex calculator should generate the same answer for the same inputs every time. There is no need for creativity, nor is there room for hallucinations. We both agreed that there was little benefit from training AI to replicate the calculator, but a conversational AI interface to gather the inputs and summarise the outputs might be helpful for some users.

The lesson is that while AI can be used for almost any use case, there are better tools for some use cases.

What is Composable Architecture?

Composable architecture refers to a design approach where individual components or modules (in this case, a combination of applications, calculators, tools, decision engines, and specialised AI agents) can be combined and reassembled to create solutions. These architectures promote flexibility, reusability, replaceability, and the rapid development of new capabilities by allowing systems to be built from a collection of interoperable and exchangeable parts.

Key Characteristics of Composable Architecture:

Modularity: Solutions are broken into discrete components, each addressing specific functions or services.
Interoperability: Each component can interact with others through well-defined interfaces or APIs.
Reusability: Components can be reused in different contexts or solutions, improving efficiency.
Replaceability: If an in-house or vendor-provided service fails to meet business or technical requirements, it can be replaced by another commercially available solution or a new custom-built in-house component.
Scalability: As needs grow, components can be scaled or replaced without affecting the entire system.

In the context of AI, composable architecture allows businesses to use specialised agents, each designed for a particular task, to form a cohesive and highly adaptive solution.

Specialised AI Agents: The Building Blocks

AI agents are autonomous systems designed to perform specific tasks by learning from data, reasoning, and interacting with their environment. When created as specialised agents, they focus on a narrow scope of expertise, such as customer service for an online retailer, skin cancer mole screening using image recognition, credit card fraud detection, or product recommendation engine.

Using OpenAI as an example, here is their list of AI products.

The point is that, even when working with a single AI vendor, models are best suited to perform the work they were designed and trained to perform.

Integrating AI Agents into a Composable Solution Architecture

Conceptually, we train a team of AI experts, each an expert in a different topic, task, or activity using the AI model best suited to their purpose. The experts will sit in the background. The user will interact with a single AI “Companion” Agent (the front end) through the user journey. The journey companion acts as an intermediary between the user, the specialists and the general knowledge base.

Specialised AI agents become the individual components or modules that are orchestrated together to form a complete solution that cohesively addresses multiple business challenges.

Key Considerations for Integration:

Interoperability: Ensure that each AI agent can communicate and interact with others via APIs. This allows them to share data and insights, enabling more complex workflows.
Data Flow and Management: AI agents depend on large volumes of data for training and operation. A robust data pipeline and governance framework must be in place to feed each agent relevant, high-quality data and metadata.
Modularity and Plug-and-Play Design: One significant benefit of a composable architecture is the ability to quickly replace or upgrade individual agents without impacting the rest of the system. For example, if a better NLP model becomes available, it can be swapped out while other agents continue functioning.
Orchestration and Coordination: A central orchestration layer may be needed to manage the interactions between AI agents. This layer ensures that agents are invoked in the proper sequence, data flows correctly between them, and their outputs are combined into a coherent result.
Scalability and Flexibility: AI agents should be designed with scalability in mind, allowing them to handle increasing volumes of data or users. Cloud-native designs and serverless architectures can help achieve this.

The Future of Composable AI Architectures

As AI continues to evolve, the trend toward modular, composable systems will only strengthen. Key innovations will likely focus on improving the seamless interoperability of AI agents and creating more sophisticated orchestration systems to manage complex real-time interactions between agents.

Emerging Trends:

AI as a Service (AIaaS): Cloud providers offer AI models and agents as modular services that can be easily integrated into existing platforms. This trend supports the composable approach by providing pre-built components for rapid deployment.
Federated Learning: This allows AI agents to learn and improve from decentralised data sources without centralising data, enhancing privacy and efficiency in modular architectures.
Explainability: As AI agents become more autonomous, it will be crucial to ensure their decisions are understandable and transparent, especially when they are used in critical systems like finance or healthcare.
Alignment: Intelligent LLM monitoring for LLMs ensures continued alignment with business goals and reduces hallucinations. Companies can operationalise their policies and known truths to help ensure accuracy and compliance with government regulations.

How this applies to my business

Illustrative Composable AI Solution Achitecture

We plan to use a composable solution architecture that leverages specialised AI agents, where they are the most cost-effective way to achieve an excellent user experience quickly, cost-effectively, and at scale. These agents will interact with knowledge bases, datasets, calculators, and tools to guide users quickly through the property ownership journey. We can reduce unhelpful variability in AI response, latency, and cost by only using AI for the tasks it does best. By breaking down capabilities into modular, reusable components, we hope to build a more intelligent, more flexible platform that can grow and evolve with our needs and our user needs and doesn’t make a pivot catastrophic. This shift enhances system agility and accelerates innovation by enabling rapid prototyping, development, and deployment of AI-driven solutions.

Through careful orchestration and integration of these agents, organizations can unlock AI's full potential while maintaining the flexibility and scalability needed to thrive in an ever-changing digital landscape.

Point-of-View: Composable Architecture Solutions with Specialized AI Agents