Unlocking AI’s full potential

Introducing AI without changing the status quo delivers limited results

So, what’s going wrong? Simply put: businesses keep relying on people-centric processes and systems. Agentic AI is not just a tool, it’s a new kind of application designed to autonomously operate and integrate business systems and tools. Without rethinking and redesigning the underlying processes and systems, introducing AI agents will never deliver the transformative results companies are aiming for.

Below, we outline six major bottlenecks and the key considerations for addressing them:

●Business and administrative processes
It is essential to evaluate the suitability of processes based on AI’s strengths and limitations. AI excels at highly repetitive, rule-based tasks but is less suited for ambiguous judgments or complex decision-making. Moreover, tasks subject to strict legal or compliance requirements, or those involving ultimate accountability, cannot realistically be entrusted entirely to AI. A thorough review and reassessment of business processes, clear definition of roles between humans and AI, and phased implementation are critical. Similar to the evolution of autonomous driving-from assistance to semi-automation to full autonomy-AI agent deployment should begin by supporting humans, gradually expanding the scope of automation. This staged approach is the fastest route to maximizing AI’s potential.

●Data
AI alone cannot reinterpret data. To leverage AI effectively, systems and tools that support business processes must be interconnected. This requires ensuring data quality, which includes consistency, completeness, and freshness. The data necessary for smooth end-to-end workflows must also be identified. If data quality issues exist care, must be taken to standardize input values, perform cleansing, and harmonize meaning through transformation processes. Companies must establish governance to manage the 80–90% of enterprise data that is unstructured and ensure real-time data utilization. Executing these steps systematically is of critical importance.

●Integration with existing systems
If AI agents are expected to operate via APIs, traditional systems designed for human screen-based operations will not suffice. It is necessary to examine and optimize the granularity of business processes and associated system transactions. API design must maintain data consistency while enabling efficient AI agent processing. This approach minimizes human error risks while preserving data integrity. AI utilization also increases the load on existing systems, making reassessment of non-functional requirements essential, particularly response performance, scalability, security, and operational stability. Completely replacing existing systems is unrealistic. Instead, retain what can be leveraged while aligning new processes with end-to-end optimization. This is key to unlocking the true value of AI technology.

●Solution architecture
Flexibility, not rigidity, is paramount. AI-related technologies, including LLMs, are evolving rapidly. Architectures must be designed to incorporate these advancements seamlessly and adapt to ongoing changes. At the same time, external LLMs frequently undergo version upgrades, so mechanisms to maintain output consistency are critical. Rather than relying solely on AI, combine it with diverse automation tools to identify optimal application areas for each technology and build an efficient, resilient environment. Structuring the architecture with loosely coupled layers and functions ensures adaptability to future technological shifts and changing business requirements.

●Management and governance
In an environment where AI operates autonomously, continuous monitoring of its validity and effectiveness is indispensable. Implement mechanisms such as detailed log collection for verifying automated workflows, real-time alerts for anomaly detection, and fail-safe shutdown capabilities when issues arise. AI agents, like humans, require clearly defined roles and escalation flows for ambiguous situations or abnormal events. Governance must extend beyond the visible front end of automated processes to include the back end of policies, rules, and audit frameworks. Ideally, these should be automated as well. This ensures controlled, transparent operations even in highly autonomous environments.

●Human resource development and operations
Agentic AI compels a redesign of human roles, from operators to managers or decision-makers. Clear role allocation between humans and AI is essential, along with training personnel to effectively utilize AI agents. Establish guidelines for handling exceptions in automated processes and ensure humans can intervene and make decisions at the right time. Beyond initial deployment, continuous improvement cycles and adaptability to uncertainty are vital for maintaining process quality, boosting productivity, and maximizing organizational performance.

What should companies do to navigate through the six bottlenecks discussed in the previous section and promote the organizational transformation required for AI adoption? In this section, we focus on three areas—business environment, architecture, and security—and present specific perspectives for each.

Redesigning the business environment

1. Designing human–AI collaborative roles and organizations
In the near future, AI agents are expected to replace many application operations currently performed by humans. To prepare for this, clearly define what tasks should be delegated to AI and what should remain human responsibilities, based on organizational and business needs. AI should be positioned as a partner that augments human capabilities.

Building on this, the goal is to have humans and AI operate as one team, not in an abstract sense, but in the way work actually gets done. That means defining which tasks AI handles end-to-end, which decisions require human judgment, and where people step in to review or approve AI-driven work. It also means giving both sides access to the same real-time information so AI can act based on the latest context and people can immediately understand why AI made a recommendation or took an action. As work progresses, humans and AI should be in constant interaction: AI presenting next steps, generating options, and executing routine tasks, while humans focus on oversight, adjustments, and complex decision-making. When we remove ambiguity about “who does what” and build processes with this partnership in mind, organizations can boost productivity and create new value that neither humans nor AI could achieve alone.

2. Simplifying and decoupling business processes
To take advantage of AI requires an event driven structure, where the completion of one task automatically triggers the next. Traditional workflows often require explicit human intervention, such as: “Person A prepares a report and emails it to Person B; Person B reviews it and requests approval from Person C.”

In an event-driven model, the sequence becomes: Report completed → system automatically notifies Person B → approval completed → system automatically sends approval request to Person C. Each process autonomously initiates based on specific events, minimizing human involvement and enabling AI agents to determine next actions more effectively. Furthermore, by clarifying standards for each task and designing processes with automation in mind, AI can operate autonomously, boosting efficiency and flexibility.

3. Real-time data collection and integration
This goes beyond simply gathering data. It means continuously collecting information from across the organization, bringing different data types into consistent formats, and making them immediately usable for AI agents. Rather than being a large, one-off undertaking, this is an ongoing process of improving how data flows through the organization. Over time, as teams refine how information is captured, organized, and shared, AI and people can operate with the same understanding and make decisions based on the most up-to-date information. This foundation is essential for humans and AI to share a common understanding.

Unstructured data such as documents, voice recordings of customer interactions, and chat logs, constitutes the majority of enterprise data and is inherently difficult for AI to use. By using technology to extract meaning-such as identifying key themes, structuring information, and filtering out irrelevant details-and organizing that information into formats AI can interpret, the data becomes usable knowledge. With this clarified and structured input, AI agents can deliver sharper decisions, stronger insights, and more meaningful support for human work.

4. Event-driven system architecture
Individual applications should autonomously interact based on specific events, enabling end-to-end processing across systems. Human roles will shift from direct “operation” to “verification and approval” of AI-executed tasks. As a result, AI agents can drive processes quickly and efficiently without delays, making organizations more resilient and adaptable.

Redesigning architecture: Progressive and continuous advancement

Redesigning architecture starts with a clear blueprint and a commitment to steady execution that combines AI with other automation solutions. Striving for perfection from the outset is not the best course. Instead, aim for incremental and continuous enhancement by business domain, which is ultimately the most efficient route to the desirable future (see Figure 3).

A comprehensive plan defines a concrete vision of the future system with clear targets, not vague ideals. This requires deep analysis of current operations to assess expected impact (assess　benefits) and implementation difficulty (assess hurdles) for each domain.

Based on these assessments, companies should develop a detailed roadmap to reach the target state, including anticipated bottlenecks and mitigation strategies. Viewing the entire landscape and setting a strategic direction minimizes rework and establishes a solid foundation for effective redesign.

Iterative improvement means creating an iteration plan based on a roadmap. Small but specific improvement cycles need to be consistently executed. Since AI and related technologies evolve rapidly, organizations must maintain a flexible approach, continuously replacing or improving systems and processes in line with the latest trends. Rather than attempting to a sudden sea change, it’s better to progress step by step, validating results along the way. This approach reduces risk while steadily moving toward the target architecture.

In short: Comprehensive planning draws the vision and roadmap, while iterative improvement drives real-world evolution. Together, they create a cycle that leads to a robust, adaptable architecture capable of fully leveraging AI agents.

I advise structuring the capabilities required to create a loosely coupled architecture that prioritizes functional flexibility and performance scalability into the following six distinct layers:

a. Collaboration and user interface layer: An interactive console that enables AI to communicate and collaborate with humans.

b. Microservices management layer: Functions that decouple and manage individual services where AI is implemented.

c. AI orchestration layer: Workflow capabilities that integrate and control multiple AI functions to execute complex, end-to-end tasks.

d. Individual functional components (AI microservices): Placement of discrete functions such as business operations, data provisioning, and governance.

e. AI core services layer: Functions that summon external and internal AI services, including LLMs and other advanced AI capabilities.

f. Operations management and governance layer: Functions for managing the entire AI system, monitoring predefined KPIs, handling escalations, and ensuring security.

AI technology is evolving at an accelerating pace. By the time implementation is complete, development of new features may already be underway. To avoid obsolescence and incomplete adaptation, each layer should be designed for modular recombination and replacement, allowing incremental enhancements as needed. This approach ensures sustainable AI utilization and contributes to long-term improvement of corporate value.

Redesigning security: Closing critical gaps

To fully take advantage of the power of AI requires businesses to rethink their security strategies . As businesses integrate diverse technologies into their environments and architectures, the risk of malicious cyberattacks increases. During this transitional phase, it is vital to anticipate AI-era threats and strengthen resilience beyond traditional security measures.

In a world where attackers routinely use AI, defenders must also fight back with AI (see Figure 5). Malicious AI-driven attacks such as phishing, malware creation and manipulation, and internal data harvesting, are becoming broader, more precise, and more sophisticated. Once inside, AI can instantly search millions of files and locate passwords with ease.

Defenders must leverage AI for security monitoring, detection, incident response, and threat hunting, while reinforcing a mindset and framework that assumes breaches will occur.

Foundational measures include implementing CSF (Cybersecurity Framework), deploying solutions like EDR (Endpoint Detection and Response), conducting vulnerability assessments, and establishing SOC (Security Operations Center) and CSIRT (Computer Security Incident Response Team) for 24/7 monitoring. On top of these, organizations must layer AI-driven defenses through comprehensive redesign.

Drawing on experience as white-hat hackers（*2）across more than 200 companies, Uvance Wayfinder consultants have identified a common trend: “Strong perimeter defense, weak post-intrusion resilience.” While traditional security measures are in place, training and frameworks assuming breaches are insufficient. Red-team（*3）test results speak for themselves:

Closing these critical gaps is the key to security transformation. Organizations must identify where the biggest vulnerabilities lie since intruders can’t attack what they can’t find. If domain admin access is blocked, they will attempt to access privileges as application admin, then database admin, and so on. Detecting and intercepting attackers during this process is an effective defense strategy.

Uvance Wayfinders provides hands-on security consulting led by white-hat hackers. Closing major gaps is nearly impossible with internal resources alone. New evaluation criteria and countermeasures informed by an attacker’s perspective are essential. By sharing Fujitsu’s practical expertise, we help customers redesign security for the AI era, from planning and implementation to continuous improvement.