Enterprise interest in AI agents and agentic AI workflows has grown dramatically in the past year. Gartner predicts 33% of enterprise apps will include agentic AI by 2028, up from less than 1% in 2024. The pace of adoption is also dramatically increasing. The number of enterprises with agentic AI pilots nearly doubled in a single quarter, from 37% in Q4 2024 to 65% in Q1 2025.
However, full deployment remains stagnant at 11% as enterprises face significant challenges implementing agentic AI workflows.
Some of the unfulfilled promises around AI agents speak to these challenges. Apple is facing lawsuits after admitting Siri would not include Apple Intelligence-enhanced features until much later than advertised. Amazon Alexa+ also recently launched with far fewer features than originally promised.
Why the disconnect? What are the main challenges in implementing agentic AI workflows that are slowing enterprise adoption?
Main Challenges Implementing Agentic AI Workflows
Enterprises face three main challenges implementing agentic AI workflows: complex system integration, stringent access control and security requirements, and inadequate infrastructure readiness. These challenges significantly slow deployment, keeping full enterprise adoption of AI agent workflows stagnant at only 11% despite rapidly increasing interest and pilot projects.
System Integration
AI-driven workflow agents need to seamlessly interact with dozens of tools, APIs, and legacy systems. Many of these systems lack the necessary APIs or aren’t AI agent-friendly.
Legacy systems in particular often lack modern, accessible APIs, which can complicate the use of agentic AI workflows designed to operate autonomously across the organization. Because these systems weren’t built with AI workflow automation in mind, AI integration can be complex and resource-intensive. Most companies require middleware, orchestration platforms, or significant development projects to seamlessly connect agentic AI workflows to business processes.
Model content protocol (MCP) is the front-runner for becoming the standard framework for orchestrating and integrating AI agents. The protocol, developed by AI vendor Anthropic, is currently supported by Microsoft, OpenAI, and Amazon, among others. There are dozens of frameworks that fit this need, so the industry hasn’t settled on a standard, but MCP currently has the most support.
This is a challenge every enterprise is dealing with right now. McKinsey's research reveals that while nearly all companies are investing in AI, only 1% of leaders describe their companies as "mature" in AI deployment.
This level of integration isn’t a “nice to have” either. Without the ability to fully navigate your organization’s systems, AI agents remain siloed and their use cases are drastically reduced.
Access Control and Security
While some systems and software can’t yet accommodate AI agents, many don’t want to allow agent access. AI agents need full access and autonomy to deliver their full promise, but many enterprises are understandably cautious about granting broad access to sensitive data and critical systems. This is a particular concern for enterprise AI use cases where authentication would be required, such as accessing data or tools that work at a user level.
Reuters Institute found 48% of widely used news websites were blocking OpenAI’s crawlers. Last year, Cloudflare released tools to allow websites to easily block AI agents from accessing their sites. Most major platforms and ecosystems, like Apple and Salesforce, limit API access and third-party agents to maintain stability and security.
We need to develop more stringent access controls and robust security frameworks before we can fully implement agentic AI workflows. This requires policy management, audit logging, and human-in-the-loop checkpoints to ensure AI agents operate within predefined parameters.
Gartner emphasizes the importance of AI governance platforms to monitor and control agent behavior, predicting that companies with robust governance will experience 40% fewer ethical incidents by 2028.
Agent Operations
Implementing agentic AI workflows also requires significant agent ops readiness, an area where many enterprises fall short. Agents require considerable computing resources, reliable, low-latency access to multiple tools and APIs, and persistent memory to maintain context throughout their operations.
Currently, there are few established frameworks or best practices for debugging, testing, and validating AI agents at scale. Unlike simpler generative AI models, agentic workflows demand continuous performance monitoring and the ability to rapidly diagnose and fix issues as they occur. Enterprises lacking DevOps-style monitoring tools tailored for autonomous agents will struggle to maintain operational stability.
The cloud architectures and performance tooling needed to run continuous, resource-intensive agent workflows aren't in place at many organizations. Without substantial investments in agent operations, the performance, reliability, and ultimately the ROI of AI agents remain limited.
Data maturity is also a major limitation. Agentic AI agents rely on accurate, structured, and accessible data to make decisions across workflows. Yet many enterprises still struggle with siloed data, missing metadata, or outdated records. Without unified data pipelines and governance, agents are more likely to hallucinate, misfire, or require human intervention.
Lenovo is solving its growing infrastructure challenges by using AI agents to streamline IT operations. Its IT infrastructure spans 23 data centers across four continents. According to Lin Qiyu, Director of Lenovo IT Operations and Maintenance Management, “Lenovo’s hybrid cloud landscape is only getting larger and more complex, which puts great pressure on our team. By using GenAI to rise to an even higher level of automation and intelligence in ITOps, we can keep up with growing complexity without an exponential increase in headcount.”
Is your organization ready for AI?
Laying the Foundation for Agentic AI Workflows
New research suggests that the time horizons for AI agents are doubling every 4 months. The performance of agentic automation in AI is ready. Enterprises are not.
The main challenges in implementing agentic AI workflows aren’t the capabilities of the agents themselves; they’re the readiness of enterprises. As powerful as AI agents are, they can’t perform effectively in environments with disconnected systems, inadequate data quality, and insufficient infrastructure.
For CIOs and CTOs, 2025 is the year to lay the groundwork for agentic AI workflows. Before AI-driven workflow agents can provide meaningful value, enterprises must achieve greater data maturity, improve integrations, enhance security frameworks, and upgrade their infrastructure.
If you aren’t part of the 65% of enterprises that have already started piloting an agentic AI project, you might not be as far behind as you think. How many of those enterprises lack the infrastructure to overcome the challenges in implementing agentic AI workflows? How many of those pilots will fail to reach production due to those barriers? By focusing on your data and AI readiness, you can lay the groundwork to ensure your organization can fully embrace AI agents.
Are you ready to assess your organization's readiness and start building the essential foundations for agentic AI? Join our AI Strategy Workshop to evaluate your current data maturity, integration capabilities, and infrastructure readiness, setting your enterprise on a clear path to scalable, secure, and effective AI agent deployment. Contact us today to reserve your spot.
