rethinking-enterprise-architecture-for-the-agentic-era_final

McKinsey Technology Practice Rethinking enterprise architecture for the agentic era Tech leaders face a choice in modernizing enterprise IT architecture: incremental change or full-scale transformation. Here are the benefits and pitfalls of each path. by Bjørnar Jensen, Florian Bauer, and Lars Vinter with Mallika Vora March 2026 -- 1 of 7 -- A company’s enterprise architecture is its blueprint for long-term value creation. Grounded in business strategy, a strong enterprise architecture comprises all the technology—networks, hardware, software, systems, and services—that makes executing that strategy possible. For decades, CIOs and CTOs have built their companies’ architectures brick by brick, integrating new elements as business goals evolve. Now tech leaders stand on a precipice: Agentic AI is changing the very fundamentals of architecture modernization. Technology leaders must quickly decide how to incorporate agentic AI into their architectures, with little precedent to guide them. They have two main choices: incremental integration, which entails deliberately adding agentic AI into existing systems to update an enterprise architecture over time; or comprehensive transformation, which requires a complete, organic overhaul of an enterprise architecture to support agentic workflows. But technology leaders will need to move fast. That’s because agentic AI is accelerating at lightning speed, collapsing traditional IT planning horizons. Tech leaders accustomed to thinking in three-to-five-year cycles must now make foundational choices in months, not years. Incremental integration allows companies to quickly deploy agents into the tech stack, but this piecemeal approach can increase technical debt and ultimately slow progress. On the other hand, a full transformation sets companies up for long-term success, but the extended implementation required could put them at a short-term disadvantage. Whichever path they choose, agentic tools can be used to speed modernization with lower risk and lower long-term run costs. Of course, choosing one path or the other is rarely a binary decision. Most organizations will pursue a middle path, which might look like domain-based modernization. That’s because few companies will find the funding to do an enterprise-wide transformation in one go. Every technological shift, whether incremental or transformational, will need to be aligned to tangible business benefits; such is the new economics of enterprise technology in the AI age. Nonetheless, examining these two paths is a valuable exercise for leaders who are exploring how to update their companies’ technology organizations. The incremental path: Building on existing foundations For many large organizations, especially those with complex legacy systems built over decades, the idea of ripping out their technology foundations overnight is unrealistic. Their data is embedded in mainframes, their business logic written in legacy code, and their processes tuned through years of iteration. For these enterprises, an incremental path offers a way forward. When it comes to integrating agentic AI, the incremental approach is rooted in pragmatism. It sees agentic AI not as a wholesale substitute for existing systems but as a layer of augmentation that can supercharge what already works. Just as the adoption of microservices enabled agile software development without dismantling the enterprise core, the first wave of agentic AI will sit atop legacy systems to extend existing capabilities. Rethinking enterprise architecture for the agentic era 2 -- 2 of 7 -- Companies can start by embedding AI agents into high-value workflows that have the ability to become automated, such as customer service, supply chain forecasting, or product life cycle management. First movers have already seen success with this approach. Each insertion brings efficiency gains, new data flows, and lessons that feed into the next stage of modernization. Over time, the hope is that these pockets of intelligence could coalesce into a more native- agent architecture. Unlocking institutional memory The incremental path does increase technical debt, but it also leverages the institutional memory of the enterprise, the processes and tech equity that have been built over time. Decades of business rules, data models, and domain expertise live within tech architectures, and agentic AI can unlock new knowledge from these assets. Replacing legacy systems is risky and expensive. Enhancing them, on the other hand, allows organizations to capture new value from old assets. An insurance company, for example, could deploy an underwriting agent that queries a legacy risk engine through APIs, translating its outputs into natural language explanations for underwriters or regulators. The underlying system remains intact, but its usability, transparency, and speed are transformed. A key enabler of this evolutionary path is the agentic mesh, an orchestration layer that connects new AI agents to one another and to traditional systems. Think of it as the nervous system that gives coherence to an otherwise sprawling digital organism. Without such a mesh, incremental modernization risks devolving into chaos. Dozens of agents, each with their own objective function, could create friction and contradiction: one optimizing inventory levels for cost savings, another for customer satisfaction, for example. The agentic mesh prevents that fragmentation by acting as a coordination fabric, enforcing business rules and maintaining a shared source of truth. The mesh also supports governance and compliance, ensuring that AI-driven decisions adhere to corporate policies and regulatory requirements. For incremental adopters, this layer is indispensable; it provides order and trust in a hybrid world where old and new systems coexist. Balancing costs and capabilities Incremental integration distributes agentic AI investment over time, allowing organizations to learn as they scale. Full-scale transformations, meanwhile, demand enormous compute power and specialized AI engineering talent, both of which are scarce and costly. Thus, companies that take an incremental approach can instead reskill and redeploy existing talent, giving employees gen AI superpowers. For example, developers could learn to build prompt-based workflows and data engineers could become AI operations specialists. This same spirit of redeployment can be applied to legacy applications. Decades-old mainframes still process a vast amount of business operations, including the majority of global financial transactions. The same will be true of today’s ERP and CRM systems. Even in the agentic AI age, they will persist. The incremental path accepts this reality. Rather than tearing down what exists, it focuses on reducing technical clutter, or the thousands of micro- Rethinking enterprise architecture for the agentic era 3 -- 3 of 7 -- applications that have accumulated over time. These can be replaced by lightweight, agent- driven workflows that achieve the same outcomes with greater flexibility. Thus, the enterprise architecture becomes more composable: a set of interoperable building blocks that can evolve with business needs. Governance as a guardrail A thoughtful incremental approach embeds governance from the start, ensuring that AI deployments operate within clearly defined ethical, operational, and compliance boundaries. Here again, the agentic mesh plays a central role by enabling centralized visibility across distributed systems. This makes it possible to audit agent behavior and enforce consistent rules. A robust governance framework is like a seatbelt in a racecar, allowing users to freely experiment with agentic AI without fear of security risks. The incremental path is less about flashy reinvention and more about architectural endurance. But technology leaders will have to carefully integrate each agentic AI deployment into their tech stacks to add intelligence without increasing technical debt. Just bolting on gen AI won’t generate real enterprise value. The incremental approach is often the smarter bet for large, risk- sensitive organizations. It preserves continuity, manages cost, and allows leaders to scale agentic AI at a measured pace (see sidebar “A case study in incremental change”). A case study in incremental change A European bank provides an example of how the incremental approach can deliver measurable value from embedding agentic AI into operations. The bank took stock of its current technology trajectory and realized that agentic AI could accelerate its innovation pathway. Working with McKinsey, the bank fast-tracked its buildup of agentic capabilities by deploying an AI-mesh architecture and robust upgrades to its tech stack. As a first step, the bank deployed agentic solutions to automate multiple types of analyses, such as financial and business risk processes, which are part of the corporate credit application process. This provided early tangible results and conviction in the effectiveness of the technology; the bank reduced time spent on administration and manual tasks. The bank then built on these learnings to scale AI agents into more workflows. Each new deployment was supported by reusable agent libraries from previously deployed solutions. This continuously expanded the bank’s repository of “atomic” agents, which execute tasks that have broad business applicability, such as data extraction. The bank created a robust security and governance framework—monitoring, maintaining, and evaluating agents continuously to ensure accuracy. The bank has now deployed multiple atomic agents and more than ten task- and context-specific agents, some of which are planning and oversight agents that monitor the other agents’ processes. The bank’s agents read various data sources, including structured financial data and unstructured qualitative meeting minutes, to produce standardized, replicable, and high-quality outputs. Rethinking enterprise architecture for the agentic era 4 -- 4 of 7 -- The comprehensive path: Embracing transformation If the incremental approach is evolution, the comprehensive approach is revolution. It calls for reimagining enterprise architecture from the ground up by placing agentic AI not at the periphery but at the core of operations. On this path, agentic AI doesn’t supplement existing systems—it replaces them entirely. Agents become the primary executors of business logic, the connectors of data, and the interpreters of intent. The enterprise evolves from a collection of fixed applications into a living network of intelligent agents capable of self-organization and continual adaptation. Unlike traditional microservices, which rely on APIs and predefined interfaces, agentic architectures are designed to be flexible and malleable. Agents can ingest unstructured data, negotiate access to resources, and modify workflows dynamically. The result is an IT ecosystem that evolves in real time, aligning itself continuously with shifting business priorities. For organizations unburdened by deep legacy constraints, an agentic AI transformation holds a strategic upside. It enables them to achieve, within three to five years, what might otherwise take a decade: the creation of a truly adaptive enterprise. Once the transformation is complete, the marginal cost of building new applications plummets and innovation accelerates. Simplified governance The irony of radical transformation is that it can simplify governance. By consolidating thousands of brittle connections into one standardized agentic framework, companies can monitor and govern an enterprise architecture more effectively than in a patchwork scenario. Yet governance must be transformed as well. The old model of static controls won’t work for dynamic agents that are continually learning and adapting. Instead, organizations can invest in AI governance platforms that monitor, validate, and coordinate agent behavior in real time. Done well, this kind of governance accelerates innovation because agents are given free rein within specific, human-monitored guardrails. This could enable, for example, much faster software development cycles. A project that once took 100 engineers a full year could be completed by a handful of teams working in concert with agent factories, such as collections of agents specializing in architecture design, documentation, testing, and deployment. Human–machine interfaces Transformation doesn’t stop at the back end. The way people interact with systems must also change, creating a fully agentic organization. For example, instead of navigating screens and forms, users will converse with digital “chiefs of staff” that anticipate needs, synthesize data, and execute actions. Such human–machine symbiosis could unlock huge productivity and efficiency gains, but only if people are supported to adapt through thoughtful change management. Rethinking enterprise architecture for the agentic era 5 -- 5 of 7 -- The comprehensive path requires a large up-front investment of time and financial resources. But it also offers the chance for companies to become first movers in agentic AI, gaining competitive advantage over companies that choose the safer incremental approach. Of course, large-scale transformations also carry risks, namely that they end up costing far more, or taking much longer to complete than expected. Our experience has shown that the majority of transformations fail to deliver what leaders had hoped, stymied by many common frustrations. Transformations may start within the technology organization, but they also require cultural reinvention. A case study in transformation One large Latin American bank recently used the transformation model to modernize a 20-year-old corporate banking platform. With a $600 million budget, it worked with McKinsey to build an “agent factory” of more than 100 AI systems that collectively redesigned legacy code, built new interfaces, and optimized data structures. Operating within guardrails, the agents entirely transformed the company’s technology processes. The result was a 60 percent reduction in engineering time and $250 million in savings—proof that transformation, while costly up front, can yield measurable returns when executed at scale. For companies that can achieve full transformation, the benefits are substantial (see sidebar “A case study in transformation”. They emerge not with upgraded systems but with entirely new capabilities: architectures that learn, adapt, and improve continuously. This turns the technology organization from a cost center into a value creator. A future-proof road map No two enterprises will walk the same path when it comes to integrating agentic AI. Some will move incrementally and others all at once. And many will pursue a hybrid strategy—starting with augmentation but designing toward transformation. Before choosing a path, technology leaders can take stock of their goals to plan effectively for the journey. Rethinking enterprise architecture for the agentic era 6 -- 6 of 7 -- As organizations navigate the future of enterprise architecture, CIOs and CTOs can follow a three-part strategy to generate maximum value from their technology investment: — Make a deliberate choice: The most important action a technology leader can take when deciding between an incremental or transformational approach to agentic change is to simply choose. Getting to a decision will call for coordinated strategy with C-suite leaders, but once it’s made, it’s made. Technology leaders can then execute quickly, implementing the technology they need to modernize the stack and hiring or upskilling to ensure teams can deliver on the change. — Modernize with and for agents: Once a path is chosen, technology leaders can focus their attention where it matters most: on how to leverage agentic AI to modernize in future-proof ways. Technology teams can use agentic AI tools to automate workflows, streamline architecture modernization, and speed up application development. But along with building with agentic AI, they must build for agentic AI—creating a future architecture that supports agent scaling. — Prioritize business impact: Modernizing technology only for technology’s sake will never deliver maximum value. The main goal of every technology organization is to improve business outcomes. When approaching any technology modernization, companies can focus first on the domains where architecture decisions will drive the greatest competitive advantage. They can then balance ambition with practicality, applying an incremental or transformational approach with the company’s risk tolerance, resources, and strategic goals in mind. CIOs and CTOs have always walked a tightrope between stability and innovation. They are tasked with creating and evolving an enterprise architecture that is secure but also cutting edge. Agentic AI only amplifies this tension. For some companies, the path of incremental integration will offer the best balance of control and progress. For others, comprehensive transformation will be the only way to seize competitive advantage before rivals do. But the greater risk lies in hesitation. In the age of agentic AI, enterprise architectures are not merely the foundation of the business; they are the business. Starting now to define an agentic architecture is the only way to achieve long-term competitiveness. Bjørnar Jensen is a senior partner in McKinsey’s Zurich office, Florian Bauer is a senior partner in the Vienna office, Lars Vinter is a partner in the Copenhagen office, and Mallika Vora is a client capabilities senior manager in the Mumbai office. Rethinking enterprise architecture for the agentic era 7 This article was edited by Kristi Essick, an executive editor in the Bay Area office. Copyright © 2026 McKinsey & Company. All rights reserved. -- 7 of 7 --