Today, the buzz around agentic AI is almost impossible to ignore. Leaders across industries and regions have spoken widely on the disruptive power of the new technology, with Microsoft CEO Satya Nadella even claiming that agents could eventually lead to the “collapse” of traditional SaaS. The potential of agentic AI means that up to half of organizations could use it in their core business functions by the end of 2025.
But even as companies race to implement this new technology, research shows that many aren’t ready to do so securely and effectively, with 86% of companies reporting that they’ll need to upgrade their tech stack to use AI agents properly.
To realize the full potential of agentic AI, organizations and their leaders should focus on these three critical areas, all of which are instrumental to long-term, sustainable success.
1. STREAMLINE DATA ACCESS AND PERMISSIONS
As AI agents proliferate, organizations need to make sure agents don’t have access to confidential information, since this can lead to compliance breaches, accidental data disclosures, and more. To do this, some organizations are using existing identity and access management for humans and extending it to AI.
At Build 2025, for example, Microsoft even introduced Entra ID for AI agents, which treats AI agents as first-class identities that can be given access rights. This kind of tool will help enforce Zero Trust principles for AI actions (full disclosure: AvePoint is an independent software vendor of SaaS solutions to migrate, manage, and protect data in Microsoft 365).
Beyond this, organizations should implement technology to automatically tag and classify documents in their cloud environments according to sensitivity and risk. This keeps the most sensitive information away from agents, which is critical, since agents could accidentally use sensitive information in an improper context if given unfettered access.
Imagine, for example, what might happen if an AI agent got hold of Social Security numbers or payroll information—the fallout could be catastrophic. That’s why organizations need strong data governance measures that can automatically regulate the flow of highly sensitive information.
2. IMPLEMENT RESPONSIBLE AI AND SENSIBLE RISK MANAGEMENT
As we saw this summer with xAI chatbot Grok’s defamatory and offensive comments, AI needs to have clear and sensible restraints in place to limit organizational risk. This, again, is true of AI assistants like Google Gemini or Microsoft 365 Copilot, but it’s particularly true with agentic AI.
Agents are useful because they’re independent, but their independence also poses risks that organizations must manage. Because they have the power to make their own decisions, organizations need to make sure agents have restrictions that prohibit them from making destructive choices. The stakes are infinitely higher when there’s not a person around to monitor its behavior and correct every mistake.
To do this, teams need to work diligently to ensure that their AI systems align with ethical and legal standards. This can mean configuring AI systems to avoid certain content, building “red team” exercises to probe for unintended behavior, and following ethical AI principles (fairness, transparency, etc.) in your custom AI solutions. If your AI agents make autonomous decisions (like approving transactions), you should have a fail-safe or human-in-the-loop for high-stakes cases, and you should also use content filters to prevent toxic or biased outputs.
It’s important to understand that, with agentic AI, you don’t have to choose between security on the one hand and productivity on the other. When controls are implemented correctly, risk management and responsible AI frameworks improve the efficacy of AI agents by eliminating risk and improving outputs, enhancing productivity and security at the same time.
3. BUILD A CULTURE THAT SUPPORTS AND REWARDS AI USE
Finally, it’s important to note that scaling AI isn’t just a technical issue; it’s also a people problem. Even with the best possible supporting infrastructure and controls in place, you still need to mobilize your workforce to drive sustainable use of agentic AI.
To do this, invest in training so employees know how to get the best out of agentic AI, including some education on its capabilities and its limitations. There’s also a knowledge gap to address: 73% of leaders say they’re familiar with Al agents, but only 45% of employees say the same, according to Microsoft.
That gap can cause friction. As you roll out AI organization-wide, run enablement sessions, create AI “champions” in each department (in the same way that some companies have Copilot Champs communities), and foster a culture of responsible experimentation.
When people are confident and educated, adoption soars. And when adoption soars, everyone benefits.
SECURING THE AGENTIC FUTURE
Agentic AI has immense promise, but the many risks and potential pitfalls of implementation and use require a thoughtful, disciplined approach. Organizations that prioritize security and efficiency by implementing streamlined access controls, AI-first cultures, and sensible risk management will be best positioned to succeed over the long term. By strengthening these foundational pillars, leaders can unlock the transformative potential of agentic AI while minimizing risk and maximizing value.