Data, agents and governance: Why enterprise architecture needs a new playbook

In “Data trust and the evolution of enterprise analytics in the age of AI,” I addressed the foundational role of trusted data and why governance is so crucial to playing a role in establishing it. What I didn’t explore was the background role of enterprise architecture (EA) in potentially being unintentionally complicit in the problem, despite its capacity to be a bigger part of the solution and, more importantly, how it can evolve to make a difference. The EA function (usually managed by IT) has not only struggled to adapt to outcome-driven business dynamics but has also unwittingly created its own existential crisis in the 21st-century enterprise. The emergence of systems of intelligence just within the last three years accelerated the crisis and challenges the function’s value, the process and its alignment with business outcomes.

If we take a sedate and unvarnished look at the reasons why, what’s obvious is that the potential value of EA never seemed to successfully integrate process and approach from a business perspective, despite adopting methodologies that purport to do so. In my experience, it rarely works on a consistent basis for most modern enterprises with a sustainable and value-driven model. 

As a former member of EA organizations for Fortune 500 enterprises in both the consumer goods and automotive industries, I frequently found myself equally as confused and frustrated as the business stakeholders I supported. I had more questions than answers. Why the dogmatic and slavish preoccupation with frameworks like TOGAF, FEAF, Zachman and capability mapping and meta models in the absence of a clear line of sight to business outcomes and actual execution? Why a similar obsession with building fiefdoms based on patterns and guardrail enforcement over pragmatism and partnership? Why not actively align, embed and support the art of the possible directly with business units and earn the coveted “seat at the table” with practical and measurable business success stories connected to the realities of the business itself? 

Clearly, systems of intelligence have amplified just how much EA has lost its way and seems hopelessly immobilized by the desire for strategic intent through dogmas, patterns and frameworks at the expense of business outcomes. The oft-quoted management maven Peter Drucker once stated, “The greatest danger in times of turbulence is not the turbulence; it is to act with yesterday’s logic.” We need a new playbook that changes the game and acknowledges three core tenets: 

1. Delivery over dogmatic approaches. When architecture principles become inflexible mandates, they quash creativity and responsiveness. 
2. Pragmatism over patterns. Indexing too heavily on “best practices” can prevent teams from adapting to unique business contexts or emerging technologies. 
3. Flexibility over frameworks. The preoccupation with framework alignment often leads to immobility, with EA teams unable to pivot quickly as new opportunities or threats emerge. 

According to Gartner, “more than 80% of CEOs expect AI to contribute to top-line growth in 2025, whereas only 3% of CIOs expect the same.” Gartner further suggests that driving that potential “can be difficult if the EA practice lacks credibility.”

I’ve seen several critical EA inflection points (cloud, API-first, etc.), but this time the shift is not only real, but it is happening in real-time. EA is indeed at an existential inflection point, as are all functions touched by AI. While CIOs struggle with accelerating AI adoption, proliferating data complexities and spiraling regulatory demands, traditional EA frameworks and practices are increasingly misaligned with the speed and scale of modern business needs. The convergence of agentic AI, next-gen data architectures and agent-based governance demands a fundamental shift in how EA positions itself to create value. 

The rapid evolution of AI and data-centric technologies is forcing organizations to rethink how they structure and govern their information assets. Enterprises are increasingly moving from domain-driven data architectures — where data is owned and managed by business domains — to AI/ML-centric data models that require large-scale, cross-domain integration. Questions arise about whether this transition is compatible with traditional EA practices. The answer: While there are tensions, the shift is not fundamentally at odds with EA but rather demands a significant transformation in how EA operates. 

The collision of traditional EA with cognitive-driven data architectures

Enterprise architecture has a storied history of providing organizations with a structured methodology for aligning IT systems with business goals, focusing on standardized business processes, data governance and technology stacks. Domain-driven data architectures, such as data meshes, support this by assigning ownership of data products to business domains, enabling agility and local optimization while maintaining enterprise-wide standards for interoperability and governance. However, this approach can result in data silos and fragmented governance, making it difficult to deliver on the promise of real-time, AI-powered insights that require data to flow seamlessly across the organization. 

Emerging AI data models: New demands and friction 

AI and machine learning models demand ongoing access to large, varied and well-governed datasets. These models often require aggregating data across varied domains, crossing the boundaries set by domain-driven architectures. This is largely at odds with most EA behaviors and comes down to three key drivers:

• Decentralization. AI initiatives often need centralized data lakes, while domain-driven models emphasize decentralized ownership.
• Complex governance. Ensuring data quality, lineage and compliance becomes more challenging as data is federated across domains but consumed centrally by AI models.
• On-demand data access. AI systems require real-time data access and adaptability, which can collide with the more fixed, process-centric nature of traditional EA frameworks.

How can modern EA bridge the gap?

According to Dharani Pothula, “Enterprise architects need to establish robust data pipelines, enforce data quality standards and implement governance frameworks that allow AI to operate effectively without compromising security or compliance.” Rather than being fundamentally incompatible, the shift to AI-centric data models is sparking a transformation in EA itself by default if not by design. Leading analysts and practitioners emphasize that EA must evolve from rigid, recurring reviews and static models to a more dynamic, real-time and outcome-focused discipline. 

Foundational and adaptive architecture opportunities for EA are many, but they demand evolutionary steps, flexibility and a responsiveness less focused on rigid constructs, frameworks and organizational structures. As I mentioned earlier, the notion of embedding or federating EA directly into business functions connects the function to business realities and the art of the possible. This notion of “infused EA” means we need a truly agile variant of EA as a practice. 

• Modern EA must support both global oversight and semi and fully autonomous business domains fluidly, frameworks for data sharing, AI governance and cross-domain collaboration. 
• AI-based data governance can also automate data quality checks, metadata management and compliance monitoring, helping EA teams manage the increased complexity of AI data flows. 
• Composability and cloud-native architectures are well paired to enable a shift toward modular, API-first and AI-Ops-based cloud-native designs, which are better suited to the demands of AI and real-time analytics. The difference is observable, intelligent and dynamic enterprise architectures. 
• Adaptive architecture is no longer just an aspirational “slideware” exercise. AI enables real-time monitoring, analysis and adaptive enterprise architectures, moving away from static documentation and toward living, evolving models. 

What about agent technology and what it means for EA? Capability mapping has long factored prominently into EA in terms of strategic alignment and transformation, roadmaps, and mergers and acquisitions, to name a few. The exercises, however, can be lengthy analysis efforts involving complex, orchestrated stakeholder alignments across multiple business units. The process, tooling and outcomes are challenging at best given the demand on time, analysis, documentation and communication and stakeholder engagement. 

So, how does agentic technology change any of this? According to Sarah Shah, VP of growth and expansion strategy at Neudesic, the capacity of agents to learn through memory and outcome reinforcement enables embedded agent organizations to become intelligent and adaptive businesses. Shah suggests that “the shift from AI automation to agentic AI represents a fundamental transformation in how organizations become intelligent and dynamic.”

How about EA practices around automation? Agency also drives a fundamental shift from static and even brittle workflow automation (RPA) to real-time, cross-platform orchestration models. According to AgilePoint’s Jesse Shah, “businesses are no longer limited to static workflows tied to a single vendor ecosystem. Instead, they are adopting abstracted, composable frameworks that can integrate agents from various platforms and execute decisions across multiple systems.”

The notion of real-time adaptive orchestration means our EA pattern assumptions have fundamentally shifted from static rules and orchestrations to adaptive interactions managed between agents that now shift the focus to a need for simulations over static patterns, principles, guardrails and governance through agent-based reinforcement learning. Agents “learn” through goals, plans and memory. Ultimately, if an agent is now empowered to adapt an architecture and workflow dynamically, the architect must focus on simulating those probable actions and understanding risk and reward in an agent ecosystem. Patterns and principles now evolve to simulations and self-orchestrated outcomes with dynamic architectures. It is undeniable that agency is the most powerful change we’ve seen in some time and it shifts our focus by default to the criticality of governance and what that means in the age of cognitive architectures. 

What is the new relationship between EA and governance and how is it different?

Historically, enterprise architecture organizations manage governance through a variety of methods, including but not limited to architecture review boards designed to orchestrate oversight, decision-making and standards enforcement. These can be lengthy, time-consuming and bureaucratic efforts where EAs navigate the enterprise armed with mandates and good intentions. It is almost always aligned around policies and standards to ensure consistency, compliance, risk management and in some cases strategic alignment.

Governance in an agentic architecture flips the script for EA by shifting focus to defining the domain authority of the agent to participate in an ecosystem. That encompasses the system they can interact with, the commands they can execute, the other agents they can interact with, the cognitive models they rely on and the goals that are set for them. Ensuring agents are good corporate citizens means enterprise architects must engage with business units to set the parameters for what an agent can and cannot do on behalf of the business.

Further, the relationship and those parameters must be “tokenized” to authenticate the capacity to execute those actions. This ultimately means that the ability to execute, validate and authenticate through simulations becomes the next level or evolution for EA governance. The balance of managing risk through intentional agent capabilities and execution plans ensures a competitive edge while not reinforcing undesirable actions and outcomes.

This evolution from data mesh to what senior technology executive and thought leader Eric Broda refers to as “agentic mesh” means siloed data domains are abstracted through agency. According to Broda, “agents and owners are provided the tools to demonstrate that the agents are working as expected. Broda also uses the term ‘certification’ instead of governance.” 

The evolutionary path may not be completely certain or clear, but successful enterprises and enterprise architecture organizations are successful because they understand how to manage uncertainty and evolve as the path becomes clear. This requires EA to be adaptive and, most of all, recognize environmental and technology triggers that can help organizations be equally as adaptive and competitive in the modern landscape of cognitive architectures. 

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?