Many organizations have struggled to realize the full value of their AI investments. Now, this trend is changing. Companies are shifting from reactive, cloud-based AI to agentic AI, making ROI potential increasingly evident. A 2025 EY report notes that 64% of UK companies now let employees independently create or deploy AI agents.
Agentic models require less human input and proactively execute complex tasks. Agentic AI advances beyond traditional, narrowly scoped systems. Organizations can augment, not replace, workers by adopting these models. The UK government supports this approach and advocates human-centered strategies to scale and de-risk AI tools.
Ensuring practicality, security, and usefulness
AI agents can understand multi-step goals, plan and sequence actions, and interact with resources to autonomously achieve objectives. For example, an agentic AI system could learn your preferences, financial constraints, and priorities. It could then independently negotiate a purchase on your behalf. Companies are already adopting these systems across industries. This is changing how business leaders view enterprise and consumer AI.
To ensure AI is practical, secure, and useful, agent workflows must use real-time intelligence. Achieving this requires a hybrid AI architecture. Hybrid AI distributes workloads across devices, the edge, and the cloud. Knowledge workers manage these processes.
Why hybrid AI is a must-have for agentic AI
Agentic AI relies on context, which often includes sensitive data. This data may be personal or organizational. Processing all data in the cloud creates privacy risks. Hybrid AI processes and makes decisions on local devices or within secure environments. This approach reduces exposure and supports compliance with data sovereignty regulations.
Personalization and data privacy are closely linked. In the purchasing agent example, user preferences and constraints are essential inputs. These often include personally identifiable information (PII) that must remain private. Processing this data locally protects privacy while maintaining the personalization that makes agentic AI valuable.
Speed is equally important. Agents negotiate deals, respond to real-time sensor data, and manage dynamic workflows—they cannot wait for data to travel across networks. Delays or disruptions cause serious consequences. Hybrid AI delivers low-latency, on-device computation. This ensures smooth and immediate experiences.
Cost is another factor. Continuous cloud processing requires significant resources and incurs high expenses. Hybrid architecture lets organizations orchestrate workloads more effectively: they route routine tasks to local devices and reserve cloud resources for demanding computations.
Hybrid architecture also supports partial task execution. Agents remain functional in offline or low-connectivity situations. They resume tasks once cloud access is restored. This mix of local intelligence and cloud-scale power makes agentic AI viable in real-world operations.
Addressing implementation challenges
Before agentic AI, organizations struggled to show clear ROI from AI investments. While agents are not a cure-all, they provide a stronger path forward when applied to complete workflows. Managing end-to-end operations with agents delivers more visible and measurable returns than working on isolated tasks.
To capture these returns, organizations must overcome several key adoption barriers.
Predictability and ethics are top priorities. AI agents must act reliably and remain aligned with human values. Organizations are quickly adopting governance platforms and techniques, such as constitutional AI. These provide necessary guardrails while preserving agent autonomy.
Complexity remains a challenge. Agents must manage demanding multi-step tasks. Teams use advances in model training and best practices to improve consistency. New development frameworks help teams build predictable, robust agentic AI systems for production use.
Secure integration with tools and APIs is also important. Agents require access to multiple data sources and applications. The industry is developing protocols and standards for secure interactions. Confidential computing technologies protect sensitive data during runtime.
Integration reliability is as important as security. Agentic AI depends on real-time interaction with external software, where failures can have cascading effects. Recent advancements—such as enhanced function-calling in foundation models and interoperability frameworks—make integration simpler. For example, the Model Context Protocol (MCP) enables secure, multi-step workflows, strengthening both agent capabilities and predictability. Key takeaway: these innovations directly improve integration reliability and agent performance.
Making it real
Agentic AI is most effective where goals are dynamic, distributed, and resource-intensive. These situations exceed the capacity of any team alone. Yet they still benefit from human intelligence for the best results.
Autonomous agents manage supply chains by analyzing real-time inventory and shipment data, helping organizations prevent logistical disruptions before they escalate. Operating on edge devices, these agents coordinate with cloud-based planning systems and update routing strategies to keep data current and secure.
On factory floors, agents embedded in industrial workstations monitor sensor data and trigger maintenance protocols. They also coordinate spare parts ordering. These actions improve operational resilience and reduce costly downtime.
AI PCs with on-device agents manage individual workflows. They summarize meetings, draft content, and interact with enterprise systems. At the same time, they do not compromise personal identity or expose private data.
Key takeaways: In each use case, a knowledge worker is essential. People not only ensure the data given to agents is accurate, clean, and contextually relevant—they also provide judgment and domain expertise that models cannot fully replicate.
Building a more autonomous future
Businesses that deploy agentic AI now and train their workforce will outpace competitors. Agentic AI forms the foundation for future advances, including AI twins, but it depends on hybrid AI. This combination lets organizations deliver autonomous, useful, and safe AI systems that perform reliably.
To lead in intelligent enterprise, companies must develop technology and build the human capability to manage it. The opportunity is open to all. Acting quickly is essential to gaining the advantage.
