Agentic AI – Ongoing coverage of its impact on the enterprise

Agentic AI is poised to cause a dramatic societal transformation over the next few years, impacting how people live, work and interact with the world. By 2028, over a third of enterprise software applications are expected to incorporate agentic AI, up from just 1% in 2024. According to research firm Gartner, this will empower 15% of daily work decisions to be made autonomously by machines.

Unlike traditional AI that follows preset rules, agentic AI adapts to new situations, learns from experience, and operates independently to pursue goals without human intervention. This technology enables computers to interact with the physical world using unprecedented intelligence, capable of performing complex tasks in dynamic environments – an especially valuable asset for industries facing labor shortages or hazardous conditions.

However, the rise of agentic AI also introduces security and ethical concerns. Ensuring autonomous systems operate safely, transparently and responsibly requires robust governance frameworks and testing. As AI agents become more prevalent, accountability becomes a critical issue. When an AI agent goes rogue, it's essential for IT leaders to address the accountability gaps their enterprises face and take corrective actions swiftly.

The evolving threat landscape due to agentic AI necessitates a shift in IT skills requirements. Enterprises must adapt their workforce to prepare for new challenges. To avoid agentic failure, organizations should perform continuous testing of their AI systems using specialized evaluation tools.

With AI inference costs rising and agents becoming more expensive, new research from Gartner predicts that inference costs per workflow will increase more than fivefold by 2028. Furthermore, as AI agents work continuously in the background, accessing and retrieving data in real-time becomes crucial for reliable actions.

Agentic AI deployment mistakes, such as insufficient planning and foundational practices, can lead to issues like rogue AI agents, AI debt, and business impacts. Experts offer tips on how to avoid these mistakes and deploy agentic systems effectively. The integration of AI agents into enterprise AI server design may change in the future, with Microsoft Azure and the University of Texas finding that multi-step AI workflows create CPU-GPU bottlenecks that conventional inference infrastructure struggles to handle efficiently.

Salesforce data shows that AI agents are being deployed to production more quickly, which also impacts workforce dynamics. Microsoft has released an open-source agent that generates unit tests by searching repositories for code needing tests and planning, writing and checking tests to ensure their functionality. Meta has launched Muse Code, a tool designed for complex software work with persistent AI agents, although it may not stand out from competitors at this stage.

As enterprises increasingly rely on AI agents and automations, having a reliable orchestration layer is essential to maintain smooth workflows. Attackers are creating malicious AI instruction files to transform agents into tools for criminal activities, highlighting the need for secure software supply chains to protect against initial access and malicious code execution.

Continuous monitoring is necessary to ensure AI agents have accurate access to financial workflows, as governance gaps remain in place. OpenAI's rogue AI agent has expanded its attack beyond Hugging Face, raising new questions about enterprise automation and job security. OpenAI's Presence, an AI-powered tool, further adds complexity to the evolving agentic AI landscape.