Unit8 Technology Radar: The State of Agentic AI – Opportunities and Challenges

In 2025 agentic AI moved from impressive demos to executive boardroom priority.

What began as copilots assisting individuals has evolved into autonomous digital agents capable of perceiving, reasoning and tracking action accross enterprise systems. Unlike traditional AI that predicts or suggests, Agentic AI can act autonomously pursuing goals, orchestrating tasks and integrating deeply into business workflows.
This shift is redefining enterprise operating models, accelerating decision-making and reshaping how organizations deliver value.
Companies that embrace agentic AI are unlocking unprecedented efficiency and responsiveness while those treating it as just another software upgrade risk falling behind.

The Rise of Multi-Agent Collaboration

Multi-Agent Ecosystems

In 2025, enterprises began deploying sophisticated multi-agent ecosystems, where specialized AI agents collaborate to manage complex workflows. Platforms like CrewAI, Autogen, or Langchain are orchestrating “agent squads”, facilitating dynamic task delegation, communication, and conflict resolution.

System Integration

Foundational protocols such as the Model Context Protocol (MCP) are crucial. They emerged the “context glue” enabling agents to securely interact with APIs, enterprise data sources, transactional systems and cloud tooling. This shift made end-to-end autonomous workflows viable, not just chat interfaces.

Foundational protocols such as the Model Context Protocol (MCP) are crucial providing the necessary “context glue” for seamless interaction with external tools, APIs, and data sources. This integration allows agents to interact seamlessly with various systems, enhancing their ability to perform complex tasks.

Vertical-Specific Agents

The focus has shifted from generic AI agents to specialized solutions tailored for specific industries like finance, healthcare, and logistics. These agents combine large language models (LLMs) with proprietary knowledge and workflows, demonstrating significant and measurable business value. For instance, in healthcare, GenAI agents assist with patient scheduling and multilingual medical inquiries while ensuring compliance with regulations like HIPAA. In finance, agents are designed for fraud detection and multilingual customer support, adhering to standards like GDPR. Retail agents offer personalized shopping assistance and inventory management, while education agents facilitate student enrollment and tutoring services.

How Agentic AI can transform different business divisions?

Legal

Many legal teams face high volumes of contract reviews while managing compliance risk. Agentic AI can automate the entire contract lifecycle—from review to analysis and tracking. This leads to faster, more compliant contract processing with significantly reduced manual effort (up to 40-70% less), ultimately lowering risk and freeing legal talent to focus on higher-value activities.

Procurement

Procurement functions often struggle with manual sourcing and purchase order creation, which can slow down operations and inflate costs. By automating sourcing, vendor research, and purchase order processing, agentic AI shortens sourcing cycles by up to 40% and reduces costs by around 20%. This means more agile sourcing and improved supplier partnerships.

Operations (Data Extraction & Workflow Automation)

In many operations teams, manually extracting data from diverse documents (like invoices and forms) leads to errors and inefficiencies. AI-driven data extraction, enrichment, and classification can streamline workflows such as invoice approvals and service requests, enhancing processing speeds, reducing manual effort, and boosting accuracy.

Operations (Reporting)

When reports are generated manually, organizations often face inconsistent data and siloed information. Automated report generation powered by AI produces consistent, accurate financial summaries, sales statistics, and compliance checks while cutting down manual work. This ensures timely insights and improved decision-making across the business.

IT

IT departments frequently deal with a heavy support burden, from routine password resets to equipment/software requests. Agentic AI can handle these routine IT support requests, resulting in approximately 35% fewer support tickets and a 75% reduction in resolution time. This allows IT teams to focus on strategic initiatives rather than repetitive tasks.

Marketing

Marketing teams need to scale personalization and optimize the return on investment for campaigns in a highly competitive landscape. With dynamic personalization and automated campaign management, AI can boost campaign performance by increasing ROI (10-30%) and driving additional revenue (up to 5-15% growth). This offers a significant competitive edge in customer engagement.

Sales

Sales processes can be slowed down by inefficient prospecting methods and manual lead scoring. Autonomous prospecting and lead scoring provided by agentic AI enriches the sales pipeline and drives revenue growth – with some organizations experiencing over an 8% revenue uplift, and more than 10% improvement in sales performance for many.

Customer Service

High inquiry volumes and the need for round-the-clock support often challenge customer service teams. AI agents capable of delivering personalized responses and automating issue resolution can dramatically reduce customer wait times from hours to seconds while increasing customer engagement by around 40%. This leads to a more satisfying and efficient customer experience.

Entreprise reality: Adoption barriers remain

Governance and Control

Despite the promising advancements, agentic AI adoption faces several hurdles. Governance and control are paramount, necessitating human-in-the-loop (HITL) approach to ensure accountability and maintain ethical guardrails standards, especially in regulated environments.

Integration with Legacy Systems

Integration with legacy systems poses another challenge, as many deployments stall due to the complexity of merging cutting-edge AI with older infrastructure. Successful approaches involve re-architecting workflows around agents rather than simply “bolting on” new technology.

Many deployment stalls due to legacy infrastructure. Successful organizations redesign workflows around agents, not bolt AI onto old processes.

Pilot Paralysis

A significant number of agentic AI projects remain in the proof-of-concept phase, hindered by issues related to cost, unclear business value, or poor risk controls.

A large portion of agent initiatives remain stuck in POC phase due to:

• Unclear business value
• Insufficient risk controls
• Over-engineering early stages

Pilot AI doesn’t create enterprise value, production AI does.

Security and Auditability Concerns

Enterprises cannot scale agentic AI without trust, control and traceability. To ensure safe deployment:

• Governance is non-negotiable when agents can access sensitive data or execute real business actions.
• Granular, role-based access control (RBAC) enforces least privilege access. Each agent gets only the permissions it needs.
• End-to-end auditability is essential: every action, decision path and system interaction must be recorded.
• Comprehensive logs and traceability turn agents from black box systems into transparent, accountable digital operators.

Implementation Strategies

Scaling from prototype to enterprise system requires decisions across:

Framework or toolset for building the logic of agentic workflows:
This decision often hinges on balancing speed with control. Low-code and no-code platforms—such as Microsoft Copilot Studio or Appsmith AI—enable rapid iteration and democratize workflow creation, making them ideal for teams that seek fast prototyping with minimal coding overhead. On the other hand, pro-code solutions like Python-based frameworks or orchestration tools such as Apache Airflow and Kubernetes offer powerful capabilities for deeper integration and customization, which is crucial for complex, enterprise-wide deployments.

Secure and governed data context:
This involves establishing well-scoped data connections—ideally through dedicated management platforms or MCP servers—that provide both the necessary access and the rigour required for compliance.

Access mode for agents:
Consider how agents will execute actions safely within transactional systems, which is vital to maintain the integrity of core business processes. Centralized model access via a single gateway guarantees consistency and control over AI models deployed across diverse workflows.

Robust governance and observability practices:
Monitoring and evaluating agent activity in real time not only mitigate risks but also ensure that the system evolves in a controlled manner as business needs change.

Deployment and orchestration
Mirror the best practices seen in modern production environments. This means enabling scalability, version control, and smooth updates—essentially treating AI agents as first-class production citizens.

Buy or build

Consider three key dimensions: technical maturity, your existing technology stack, and time-to-market priorities. Many organizations find that a hybrid approach—purchasing components to accelerate deployment while custom-building parts that serve as unique differentiators—strikes the optimal balance between agility and control.

In summary, building agentic AI capabilities requires a strategic investment in the right tools, secure and governed data environments, and robust deployment processes. By carefully assessing the trade-offs between low-code, no-code, and pro-code solutions, and understanding the reasons behind each decision, C-level leaders can drive a transformation that delivers faster, safer, and more efficient business outcomes.

Architectural Considerations

Placement in Enterprise Stack

Determining the placement of AI agents within the enterprise stack is a critical architectural decision with several alternatives to consider.

There are a few distinct options:

• Edge/UI: This is the current trend exemplified by copilots and ChatGPT, where AI functionality is delivered at the user interface. It is fastest to ship, highly demo-friendly, and supports discovery and self-service. However, this approach often involves tab-hopping and copy-pasting between applications, and typically lacks true end-to-end actions and rigorous policy enforcement.

• In-App, Embedded AI: Here, AI capabilities are built directly into key enterprise applications (such as ERP, CRM, or SAP modules) and leverage native permissions for domain-deep tasks. While this integration provides robust support for individual system functions, it tends to create silos and may struggle with orchestrating cross-system processes.

• Agent Orchestrations Layer: In this option, the AI agents are positioned between the intelligence layer and the systems of record. Acting as the control plane for agents, Orchestrations Layer provides a connector directory/hub along with robust policy and governance (RBAC/ABAC, allowlists, application-specific scoping, approvals, and DLP), human-in-the-loop review for high-risk actions, and comprehensive observability and audit capabilities (including traces, evaluations, canaries, rollback measures, eval management, and sandboxing).

For mature enterprise deployments, the recommended approach is Option C. By placing agents in an Agent Orchestrations Layer, organizations ensure that agents have access to semantically rich context (capturing the enterprise ontology), respect permissions, and can execute real transactions—such as raising a purchase order, updating CRM entries, or reconciling invoices—with auditability built in by default. This placement ensures that agents have access to semantically rich context, enabling them to understand the ontology of the enterprise and execute real transactions with audit by default.

Target Architecture

An effective agentic AI architecture requires several key components: a workflow-building framework or tool, secure context access, safe action execution, centralized model access, comprehensive governance and observability, and robust deployment and orchestration capabilities. Organizations must decide whether to build or buy these components based on their technical maturity, existing technology stack, and time-to-market considerations.

Empathetic and Human-Centric AI

As agents take on more customer-facing roles, the focus will shift to making interactions more human-centric. Agents will be equipped with empathetic capabilities to understand human emotion and intent, enabling more personalized and effective customer interactions. The most successful models will not replace human workers but augment them, evolving the human role toward governance, strategy, and ethical oversight.

The Path Forward

For businesses to thrive in the agent-powered world of 2026, a proactive strategy is essential. This means investing in governance and observability frameworks to manage agent behavior effectively, building robust integration capabilities to connect new agentic systems with existing infrastructure, and prioritizing data quality and security from the outset to ensure reliable and compliant agent operation. The evolution of agentic AI promises a future where businesses are not just faster but smarter, more resilient, and more collaborative. However, navigating this transition successfully will depend on a careful balance of innovation, governance, and human-AI partnership.