Enterprise Architects: It’s time to flex your AI brain and build, not just buy, a smart copilot

This article is cross-posted on each of the co-writer blog: William El Kaim, Stéphane Goudeau and Samuel Pariente.

For years, enterprise architects have navigated a constant challenge—balancing between basic tools and specialized software while striving to keep information accurate, consistent, and up to date. Despite technological advancements, maintaining a clear and comprehensive view of the enterprise landscape has remained an uphill battle.

However, the rise of AI is set to transform the game. With its ability to seamlessly ingest both structured and unstructured data, AI can analyze an entire information system holistically, providing unprecedented insights and support. This shift marks the beginning of a new era—one where architects are no longer burdened by fragmented data and outdated models but are instead empowered by AI-driven copilots.

Architects' Copilot: Built, Not Bought

By “reshaping” their function and key tasks with AI, architects can boost their productivity, often by more than 50%. Rather than merely adding AI tools to existing workflows, this approach involves redesigning core processes and skills around fit-for-purpose technology and data platforms. ?

Interestingly, we see more and more architect publishing on these opportunities, like Jesper Lowgren (see below), Timo Elliott, or Thilo Hermann.

Unlike off-the-shelf software solutions, an architect’s copilot is not something that can simply be purchased. It must be carefully developed and nurtured within your organization, tailored to your unique processes, challenges, and goals.

Agentic AI to build reasoning copilots

In the early days, modest prototypes and basic gains were achieved using off‐the‐shelf AI tools (like chatGPT). Now, with reasoning-capable solutions available at minimal development costs (DeepSeek R1, OpenAI 03, Gemini 2.0 Flash), AI agents are not just answering queries—they’re thinking through problems step by step, gathering real‐time insights, and taking decisive actions. Agents can continuously learn from the environment and evolve over time.

The traditional architect’s way of working is now being upended as AI transitions from task automation to orchestrating complex, agent-driven processes. In this new paradigm, architects’ tasks and decisions are empowered by self-learning agents. AI “copilots” are then a set of orchestrated agents, with a unified and simplified user experience. These copilots can also be used by non-architects to foster knowledge sharing, self-assessment and best practices learning.

The 10-20-70 rule

Yet this revolution is not without its challenges. Pioneers of AI at scale typically follow the 10-20-70 rule (see image below), with 70% of their AI investment centered on embedding AI into business processes and new ways of working. Imagine an organization adopting an AI-driven decision tool for application portfolio management. If they invest 10% in pilot experiments and AI training, 20% in tool and model development, they still must devote 70% to restructuring daily workflows. This could mean new collaboration patterns between architects and data scientists, redefining governance models, or implementing continuous learning loops to keep the AI agent’s insights relevant and up-to-date.

The world of architecture is being forced to confront long-held practices. As architects navigate a rapidly evolving technical landscape, they must reconcile established methods with the need for agility and continuous learning. The very tools that enable innovation, like AI-driven decision-making frameworks, require new approaches to training and integrating knowledge across teams and systems. This challenge, though formidable, is the engine that drives the transformation. The architecture of today must evolve to not only embrace these technologies but to leverage them in ways that create long-term, scalable solutions that will respond to future complexity.

Agentic AI framework to Reason and Act (ReACT)

Agentic AI frameworks combine reasoning and action, enhancing AI’s ability to interact dynamically with external tools, adapting to evolving contexts, and continuously refining decision-making processes across diverse tasks. These frameworks, powered by large language models (LLM), enable everything from simple tool operations to more intricate workflows, ensuring flexibility in addressing complex challenges.

Among these frameworks, ReACT stands out for its seamless integration of reasoning and action in real-time.

“ReAct, short for Reasoning and Acting, is a groundbreaking approach that intertwines reasoning (the AI’s thought process) with action (the AI’s execution of tasks). Unlike traditional AI models that treat these processes separately, ReAct integrates them into a cohesive workflow. This integration ensures that every action the AI takes is informed by its reasoning, and every piece of reasoning potentially leads to a meaningful action. The result? A more adaptive, responsive, and effective AI model.” (source Jomsborg Lab).

In practice, this means an AI agent can propose a solution, query external data sources to validate it, refine the approach based on real-time feedback, and then implement recommended changes. The loop of reason–act–learn remains ongoing, continually improving the agent’s ability to handle complex architectural challenges.

Introducing Archie, our Architect copilot

Archie is a prototype of an Agentic AI copilot, created by BCG Platinion Paris. It follows a “ReACT” architecture (see figure below) and can natively connect to sources of knowledge available, like the enterprise architecture playbook we built (our bible) and the EA tool (e.g. LeanIX).

When receiving a request, Archie try to find a solution, following these steps:

·?????? The agent compares potential solutions, taps into LeanIX to see which applications will be affected, and recommends the most viable option—all while learning from your feedback.?

·?????? The agent can draw on multiple sources, monitor changes in real time, and push relevant insights or alerts directly into LeanIX.

·?????? The agent automates checks across the portfolio, flags anomalies, and suggests remediation steps?

While these steps outline Archie’s current capabilities, future iterations could push its potential even further. For example, if an organization considers retiring a critical application, a future version of Archie could not only analyze dependencies but also flag potential compliance risk, generate alternative transition plans, and trigger follow-up discussions with relevant stakeholders, thus reducing manual oversight and miscommunication when these features are fully implemented.

By integrating AI-driven decision-making and knowledge platforms like LeanIX, we can evolve static repositories into living, responsive knowledge hubs. Within a unique user interface, it is then possible to ask questions in plain English and to let the agent use the sources of data required to produce the appropriate answer.

With a ReACT agent at its core, Archie can compare potential solutions, assess the impact on application portfolios, and offer strategic recommendations, all while learning from user feedback in real time. The synergy between AI agents and traditional software platforms highlights a broader business impact: legacy SaaS models are shifting as the real “intelligence” migrates into an agile AI layer that orchestrates across systems, driving efficiency and innovation.

The future of enterprise architecture is not just about better tools—it’s about intelligent collaboration. And with AI as a copilot, architects are poised to unlock new levels of efficiency and strategic impact. Ultimately, an environment where architects and AI collaborate seamlessly can become a strategic differentiator, driving faster innovation, reducing errors, and opening doors to new business models.

#AI #Agentic #Copilot #Architect