Rewriting the rules of enterprise architecture with AI agents

In “Data, Agents and Governance: Why enterprise architecture needs a new playbook,” I examined the collision course that agentic intelligence and enterprise architecture face, and how governance automation and simulations represent the next level of evolution for enterprise architecture. Before we delve into exactly what that means, we need to step back in time to understand the power of simulation in managing higher-risk scenarios, such as challenges like supply chain optimization.

The Rand Corporation was a pivotal American think tank established in 1946 to provide research and analysis to the US military, particularly the Air Force, during the Cold War. Its influence extended far beyond technical studies, shaping US defense strategy, nuclear policy and the broader framework of Cold War deterrence.

In the late 1970s and 1980s, Rand developed advanced, computer-supported war-gaming systems. These systems used automated agents to simulate the decision-making of both superpowers (the U.S. and the Soviet Union), as well as third-party countries, through repeated, rigorous analysis of complex scenarios without humans necessarily running every iteration. The realism and scope of strategic analysis with multiple permutations, including operational constraints, alliance characteristics and the potential for mistakes or miscalculations by both sides. Simulations and game theory profoundly influenced the planning process associated with nuclear defense and deterrence, in addition to influencing decades of post-war military strategy.

So, what does the history of the Rand Corp. have in common with EA, governance and agentic AI? Rand changed how we view and manage risk with the highest geo-political stakes using deep analysis of complex scenarios without human intervention. Agentic AI is poised to do the same in the modern enterprise. The risks are not equivalent, but the conceptual playbook is fundamentally the same. In a literal sense, agents act on behalf of another party. They effectively act as a proxy. The degree of autonomy they have in executing a business process drives how agency can be implemented as a governance tool and the nature of agentic behavior introduces risk as well…which is why oversight automation (agentic governance) is so critical and one half of the EA’s future.

So, what about the other half? Given assisted, semi-autonomous and fully autonomous options, the risk is real when you consider the ability of agentic AI to use reinforcement learning, memory and goal-oriented objectives to interact with other agents within and outside of a business ecosystem. For many enterprises, this ultimately means EA can play an active and very different role in providing oversight through agentic simulations, model curation and certification, and even using agents to manage, measure and execute simulations and outcomes.

As an architect supporting an API mobile ecosystem enablement effort for a global sports apparel enterprise, I learned just how critical simulations are to telemetry and biometric wearable performance management. Simulations were equally as critical for me as a principal architect in facilitating measurable and observable outcomes in connected vehicle-connected consumer solutions in the automotive industry. Simulations support observability, and since agents “learn” and even pivot based on what they learn, agent observability is critical in understanding risks relative to the business functions we automate and the degree of autonomy we support within and outside our business ecosystem.

In cloud architectures, best practices around proscriptive environment configurations are routinely automated to ensure that critical factors, like sizing, cost, performance and security, are embedded through infrastructure-as-code by default and by design. CI-CD pipeline deployment automation enforces blue-green deployments to test assumptions, reduce risks associated with defect resolution and limit the “blast radius” of defective features through targeted and limited, soft-launch audiences. Simulation in some form is already used as a routine practice today to provide observability metrics and use those metrics to manage very specific, risk-sensitive, governed outcomes.

More sophisticated simulation models and testing, however, are also not new to IT, software engineering or enterprise software portfolios. We’ve seen attempts to manage and optimize workflows with robotic process automation, business process management and business process optimization. These sometimes fragmented and disconnected solutions mostly facilitate an understanding of current processes and models and help us optimize or refactor processes through brittle, scripted implementation.

The missing piece was not the ability to model and simulate, it was the intelligence automation to analyze, execute and adapt. Enter LLMs, agentic AI and what is now mature digital twin technology for comprehensive simulation of process, system, technologies and ecosystems. The term “digital twin” was originally used by Dr Michael Grieves at the University of Michigan in 2002, during a presentation on product lifecycle management (PLM), so while it found a home in industrial settings, the concept of a mirrored instance to validate design assumptions, risk and reward has broad applications. Further, the marriage of agentic AI with digital twin technology poses an interesting opportunity. But first, let’s look at the respective capabilities of both technologies. 

What can agentic AI do? Understanding agentic AI in the enterprise context

What are the characteristics of agentic AI?

• Autonomy: They operate autonomously or semi-autonomously, making decisions with or without continuous human intervention.
• Goal-Orientation: They follow specific objectives, adapting their strategies as situations change and they learn through adaptation, memory and reinforcement learning.
• Context awareness: They perceive and interpret their environment, using real-time data to inform actions and store experiential learning.
• Learning and adaptation: They improve their performance over time through feedback loops, memory and reinforcement.
• Collaboration: They can interact with humans, other agents and systems to coordinate actions and share knowledge. 

So, how does agency fit into EA’s future? According to the Eva Jaidan, head of artificial intelligence at MEGA in the Enterprise Architecture Professional Journal, enterprise architects are the “guardians of AI agents” and that “by mapping IT systems, processes and business capabilities, architects can determine which systems AI should connect to, where it can create the most value and how to align initiatives with the company’s strategic objective.”

In enterprise architecture, agentic AI systems can be deployed as digital “co-architects”, process optimizers, compliance monitors and scenario planners — each acting with a degree of independence and intelligence previously impossible.

So why agentic AI and simulations for governance…and why now? Governance in enterprise architecture is about ensuring that IT systems, processes and data align with business goals, comply with regulations and adapt to change. The traditional governance model relies deeply on frameworks, manual review, periodic audits and static policies. These methods are increasingly inadequate in the face of real-time business dynamics.

Agentic AI introduces a new composability model that is achievable: Governance that is continuous, adaptive and proactive. Agentic systems can monitor the enterprise landscape, simulate the impact of changes, enforce policies autonomously and even resolve conflicts or escalate issues when necessary. This results in governance that is both more robust and more responsive to business needs. Gartner’s research reinforces the impact of agency and simulations on enterprise architecture’s future. According to its Enterprise Architecture Services Predictions for 2025, 55% of EA teams will act as coordinators of autonomous governance automation by 2028 and shift from a direct oversight role to that of model curation and certification, agent simulations and oversight, and business outcome alignment with machine-led governance.

The marriage of digital twins and agentic AI: A foundation for real-time simulation

A digital twin is a dynamic, virtualized implementation of a physical asset, process or system, continuously updated with real-world data. The use, role and maturity of digital twins have evolved significantly over the last decade. According to McKinsey, the global market for digital twin technology is expected to grow 60% annually and reach $73.5 billion by 2027. Unlike traditional simulations, which rely on static or predefined data, digital twins maintain a real-time feedback loop, mirroring the current state and behavior of their physical counterparts. This capability allows organizations to:

• Test scenarios without real-world risks: Digital twins provide a safe environment to simulate process changes, regulatory adjustments or operational innovations, minimizing the risk and cost associated with physical prototyping.
• Optimize operations: By integrating live data, digital twins help identify inefficiencies, forecast outcomes and support data-driven decision-making, often resulting in measurable cost reductions and improved performance metrics.
• Enhance scenario planning: With digital twins, enterprise architects can simulate the effects of strategic decisions, regulatory changes or external disruptions, enabling better risk management and resource allocation.

For example, in the financial sector, a bank can use a digital twin to simulate the impact of new compliance requirements, testing how changes affect workflows, training needs and system dependencies before deploying them across the organization. In manufacturing, digital twins enable companies to model supply chain disruptions or process optimizations, providing insights that drive resilience and efficiency.

Agent-based models and digital twin simulation labs: A logical evolution for governance

While digital twins offer a powerful platform for simulation and analysis, the new frontier is the convergence of both agentic AI and digital twin simulation — autonomous, context-aware agents capable of reasoning, planning and acting within these virtual environments. Agentic AI moves past traditional automation models by introducing systems that can:

• Perceive and interpret their environment: Agents continuously monitor digital twin data, recognizing changes, anomalies or emerging risks in real time.
• Make autonomous decisions: Within defined ethical, legal and operational boundaries, agentic AI can self-regulate, enforce governance policies and initiate corrective actions without waiting for human intervention.
• Learn and adapt: These agents refine their strategies based on feedback and evolving data, ensuring governance remains effective as business contexts shift. 

The combined power of digital twins and agentic AI is revolutionizing enterprise governance. Digital twins provide the real-time, dynamic simulation environments necessary for understanding and optimizing complex systems, while agentic AI introduces autonomous, adaptive agents capable of managing governance at scale and speed. The marriage and opportunity for a new EA rebrand and playbook are clear:

• Integrate simulation labs (“digital twins”) with enterprise architecture: Use simulation and validation as the foundation for real-time scenario analysis, ensuring they are continuously updated with operational data.
• Define clear governance boundaries for agents: Ethical, legal and operational considerations that agentic AI must follow, with mechanisms for human oversight and escalation.
• Invest in data quality and integration: High-quality, accessible data is essential for both digital twins and agentic AI to function effectively.
• Monitor, audit and refine: Deploy monitoring tools that track agent decisions, provide audit logs and support continuous improvement of governance models.

As organizations continue to embrace this paradigm, agent-based AI models will become the heir apparent and dominant approach for future governance — delivering continuous assurance, proactive compliance and resilient, self-improving enterprise architecture.

This article was made possible by our partnership with the IASA Chief Architect Forum. The CAF’s purpose is to test, challenge and support the art and science of Business Technology Architecture and its evolution over time, as well as grow the influence and leadership of chief architects both inside and outside the profession. The CAF is a leadership community of the IASA, the leading non-profit professional association for business technology architects. 

This article is published as part of the Foundry Expert Contributor Network.
Want to join?