Conceptual Thoughts

This section provides background on the concepts behind the MAC project. Generative AI and Agents are top of mind for many organisations. As with all hypes, companies are rushing in, afraid to be left behind. Most organizations are in experimental mode, using one or more of the 3 types of frameworks that are being offered on the market to build agents. From fully packaged and managed frameworks, to completely custom code based frameworks. Almost every framework holds a promise of easily building agents by providing pre-built or standard capabilities to serve the needs of the implementation.

The result is that we are rushing into the building phase, without thinking about why we need an agent in the first place. We also miss out completely on the lifecycle phases of the agents, from design to deploy, including security and operational aspects.

And this is creating "AI Sprawl"; the uncontrolled proliferation of agents and callouts to AI providers across an organization. Not managed, this can cause some serious damage to an organization, e.g. confidential data exposed to the outside world.

This may remind us of the past, when we were starting to build APIs and Webservices. It was all about getting the work done, without taking a step back to understand other life cycle aspects like designing & prototyping to introduce agility, speed and reuse. Let us step back in history to understand how the evolution of IT has changed the way we implement and deliver software technology.

The evolution of technology and organisational setup has seen shifts across various domains, all driven by the need for greater adaptability, integration, and efficiency. This transformation has moved through several phases, each marked by significant advancements in teams, processes, architecture, and deployment models, which ultimately laid the foundation for today’s flexible, building block based, data-centric digital world.

Evolution in Key Areas

• Teams and Collaboration: Initially, teams were isolated, with business, development, and operations functioning as separate silos. Over time, as the need for faster, more integrated responses grew, teams began to work collaboratively. Today, we see cross-functional teams that bring business, development, and operations together, promoting a seamless, agile approach to product development and deployment. This also addresses ownership of digital building blocks; cross-functional, domain aligned teams will own all digital building blocks within their domains.

• Development Processes: Development began with traditional, linear methods like Waterfall, which were slow to adapt to change. The shift to Agile methodologies brought iterative development, allowing teams to respond to feedback and evolving requirements more rapidly. Today’s DevOps practices integrate development and operations to ensure continuous integration and delivery, accelerating product cycles and creating a smoother path from concept to deployment.

• Application Architecture: Applications evolved from monolithic structures, where all components were tightly coupled, to N-tier architectures that separated layers for improved scalability and manageability. The adoption of microservices architecture has further transformed application design, allowing small, independently deployable services to work together. This approach enhances flexibility, scalability, and resilience, as each service can be developed, deployed, and maintained independently as digital products.

• Infrastructure and Deployment Models: Infrastructure has shifted from on-premises data centers to hosted solutions and, finally, to the cloud. Cloud infrastructure offers unparalleled scalability, flexibility, and cost efficiency, making it a standard for modern applications. With the introduction of containerization, applications are now packaged with all dependencies, enabling them to run consistently across various environments, further enhancing deployment agility.

• Data Architecture: Data has evolved from isolated warehouses to centralized lakes that handle vast amounts of structured and unstructured information. Today, data is treated as a product in itself, with architectures like Data Mesh providing decentralized governance, enabling individual domains to manage their data independently but in a way that supports the larger organizational ecosystem.

APIs, Data, and Agents as Products

As technology matured, APIs became a productized interface for software integration, creating a seamless channel for inter-system communication. Initially, APIs were just internal tools, but they eventually grew into monetizable assets, allowing businesses to expose functionality to developers and generate new revenue streams. API-led connectivity has since become a foundation of modern integration, allowing different systems to interact and enabling developer ecosystems around companies like Google, Facebook, and Stripe.

Data followed a similar trajectory, evolving from isolated assets into valuable products. Organizations began to recognize the intrinsic value of their data, organizing, governing, and monetizing it to drive insights, decision-making, and competitive advantage. Today, data products offer contextual insights, focusing on specific business domains and serving business users directly. APIs often support data products by providing necessary access and integration.

The Next Evolution: Agents as Products
Now, the stage is set for the next phase—Agents as Products. Just as APIs and data products evolved into key business assets & enablers, agents will soon take on a similar role. Building intelligent agents is becoming accessible to anyone, but how do we govern it.

Building agents is no longer rocket science—it's the logical next step in transforming technology from isolated tools to intelligent composable building blocks that can drive business value at scale. To leverage AI agents effectively, we need to think beyond development and into lifecycle management, governance, reusability, and alignment with organizational goals.

Creating AI agents in a business context goes far beyond initial implementation. It’s about embedding your intellectual property, processes, and unique insights into a future-proof framework that empowers teams and drives operational excellence. For agents to truly thrive within an organization, they must be designed, built, and managed as products with a complete lifecycle—from Design and Prototyping to Operations and Governance—each phase meticulously handled with a holistic, long-term view.

Lifecycle Management

AI agents must be part of the organization’s value streams, aligned with strategic goals and designed to evolve alongside changing business needs. Lifecycle management is not merely a series of checkpoints but a structured process that considers everything from concept to decommissioning. Every stage—design, development, deployment, and maintenance—should align with business objectives, ensuring agents provide sustained value rather than short-term solutions.

Composability and Reuse

The real power of AI agents lies in their adaptability and modularity. By designing agents with composable and reusable components, organizations can efficiently integrate AI across various domains like HR, CRM, ERP, and beyond. Components like knowledge bases, language models, and task-specific logic can be reused across different departments, reducing redundancy and ensuring consistent intelligence across the organization. Composable agents also allow businesses to quickly iterate and adjust, responding to new demands without starting from scratch each time.

Security and Governance

As AI agents interact with sensitive data and make impactful decisions, they must be developed with stringent security and governance frameworks. Effective management of agents includes covering essentials like:

• PII Masking to protect personal data and comply with regulations.
• Toxicity Detection to ensure ethical interactions and prevent harmful outputs.
• Cost Control mechanisms to monitor and optimize usage, avoiding unforeseen expenses and waste.

Governance goes hand-in-hand with security. AI agents must be designed and operated under clear ethical guidelines and regulatory compliance, with safeguards to prevent bias, misuse, and unauthorized access. By ensuring robust governance, organizations can confidently deploy AI without compromising data integrity or ethical standards.

Insights and Analytics

To maximize the impact of AI agents, organizations need to track their performance, costs, and business alignment across Lines of Business (LoB). Real-time analytics allow organizations to measure the agent’s effectiveness, identify bottlenecks, and continuously refine their capabilities. With detailed insights, businesses can ensure that agents don’t just operate efficiently but actively contribute to strategic goals. Insightful analytics enable leaders to make data-driven decisions, refine the agent’s abilities, and align its evolution with shifting business objectives.

Domain Agents & Agent Networks

Agentic Architecture goes beyond traditional siloed systems by establishing a holistic ecosystem where AI agents function seamlessly across various high-level business domains. Imagine an enterprise as a collection of specialized domains:

• 𝐂𝐮𝐬𝐭𝐨𝐦𝐞𝐫 𝐃𝐨𝐦𝐚𝐢𝐧: In areas like sales, service, marketing, and commerce, Salesforce Agentforce agents are already revolutionizing customer interactions and enhancing experiences.
• 𝐎𝐭𝐡𝐞𝐫 𝐊𝐞𝐲 𝐃𝐨𝐦𝐚𝐢𝐧𝐬: AI agents are also primed to optimize workflows in HR, Finance & Accounting, Product Development/R&D, and Supply Chain & Operations. In each domain, AI agents perform tasks, automate processes, and make real-time decisions to drive efficiency and value.

So, how do these diverse agents work together effectively?
Enter 𝑴𝒖𝒍𝒆𝑺𝒐𝒇𝒕 𝑨𝑰 𝑪𝒉𝒂𝒊𝒏 (𝐌𝐀𝐂) 𝐏𝐫𝐨𝐣𝐞𝐜𝐭, the unification layer that enables seamless communication between AI agents, both within and outside individual business domains. MuleSoft AI Chain acts as the connective tissue, ensuring that each AI agent is empowered with the data and insights it needs, regardless of its origin or location within the enterprise.

As we delve deeper into the world of interconnected AI, it’s vital to have a platform capable of governing the lifecycle of agents as products. This involves managing each agent's development, deployment, and lifecycle, so they can evolve as the enterprise grows. Governing agents as products ensures that they can be continuously optimized, adapted, and retired as needed, reducing risk and enhancing resilience.

The goal is to empower organizations with a robust infrastructure where AI agents are not just tools, but cohesive components of a unified ecosystem. MuleSoft AI Chain supports and complements both Salesforce Agentforce agents and other third-party agents, transforming disparate systems into a coherent Agentic Architecture for the future of work.