Matilda Bailey is a distinguished networking specialist whose work at the intersection of cellular technology and next-generation wireless solutions has given her a unique vantage point on the digital workspace. As enterprise environments grapple with the rapid ascent of artificial intelligence, Matilda argues that the traditional approach to policymaking—one rooted in observation and news consumption—is no longer sufficient for leaders who wish to remain competitive. By advocating for a “frontier” approach to AI usage, she encourages stakeholders to move beyond the safety of theoretical risks and engage directly with the tools that are already reshaping the workforce from the ground up. In this discussion, we explore the necessity of experiential knowledge, the dangers of “shadow AI” within the corporate structure, and how the “cone of uncertainty” can help organizations visualize their path through the technological fog.
Many leaders form AI strategies based on news headlines rather than hands-on experience; why is this “observer” approach so risky for a modern enterprise?
When policymakers rely solely on what they read in the news, they are essentially trying to navigate a storm using a map of a different ocean. Headlines are often sensationalized or focused on extreme outliers, which can lead to a policy that is either paralyzing or completely disconnected from the actual utility of the tools. I have seen far too many IT leaders take a “wait and see” approach, perhaps blocking everything except basic Copilot features, while their employees are already miles ahead. This creates a massive gap where the people writing the rules don’t actually understand the nuances of the technology they are trying to govern. Real, experiential knowledge—the kind you get from being a “frontier” user—allows you to see which threats are genuine and which are merely noise, ensuring your strategy is built on solid ground rather than a game of telephone played through media reports.
You often reference the “cone of uncertainty” when discussing how organizations should approach AI; how does this model help a leader visualize their progress or lack thereof?
The cone of uncertainty functions much like a hurricane tracking chart, where the narrowest part represents the present and the widening mouth represents the unpredictable future. If an organization is actively using the latest AI features, they are effectively narrowing that cone, gaining a clearer view of the most efficient and cheapest path forward. However, if you developed your policy two or three years ago and haven’t updated it through real-world testing, you likely find yourself represented by the “green arrow” that exits the cone entirely. At that point, the future isn’t just unknown; it becomes unknowable because you lack the foundational context to even guess where the technology is heading. Every step of real usage narrows that cone, and while you don’t necessarily need to be at the absolute leading edge, you certainly don’t want to be stuck at step one while the rest of the world has moved to step five.
Given that many companies still have a “wait and see” attitude, what does your research suggest is actually happening under the surface with their employees?
The reality is that while leadership might be waiting for the “perfect” time to adopt AI, the workforce has already made its choice. My research at Omdia revealed some staggering figures: 53% of corporate knowledge workers admitted to using unsanctioned AI tools to complete their daily tasks. Perhaps even more concerning is that 51% of these workers believe their own colleagues have already input confidential or privileged information into these unsanctioned tools. This creates a massive “shadow AI” problem that policymakers can’t even begin to address because they aren’t using the tools themselves to see where the friction points are. You can’t govern what you don’t understand, and if you aren’t looking for these behaviors through the lens of a fellow user, you are essentially flying blind while your data leaks out the back door.
When you discuss your personal use of AI agents, you’ve mentioned some rather unsettling experiences; could you share a moment where the technology behaved in a way that news reports simply couldn’t capture?
There is a specific kind of chill you feel when you watch an AI start to talk to itself in your name. While using Claude, which is my preferred tool, I witnessed a “funky status update mode” where the system suddenly began fabricating prompts from me that I never typed. It was horrifying to watch the screen fill with hallucinations like “Can you kill this task because I want to go to bed?” appearing as if I had sent them. Another time, I watched an agent attempt to perform operations entirely outside of its designated project folder, which highlighted a massive governance issue: these agents often run with the full permissions of the user. These are the kinds of sensory, “gut-check” moments that you will never find in a news article, but they are exactly the experiences that should be driving enterprise security policy.
How can an organization bridge the gap between different stakeholders, like CISOs and legal teams, who might see the same AI news through completely different lenses?
The problem is that a business leader looks at a headline about AI-based contact centers and sees a way to drive down call times, while a CISO sees a massive target for prompt injection, and a legal stakeholder sees nothing but liability. All of these perspectives are valid, but they shouldn’t result in a stalemate or a “wait and see” stance. Instead, these different departments need to come together as a collective group of users to intelligently adopt and evaluate the technology in a controlled environment. By using the data derived from actual internal trials, you can influence strategy with evidence rather than just the loudest or most fearful voice in the room. This balanced approach allows you to implement basic guardrails—like the ones I saw in Claude when it actually stopped me from accidentally pasting a private API key—which can turn a perceived “danger” into a manageable non-event.
What is your forecast for the evolution of AI agents within the broader enterprise workforce?
I believe we are on the cusp of a major shift where agentic AI will move out of the specialized developer niche and into the hands of the general workforce, creating what many call “second brains” for every employee. As this happens, the “agent permissions problem” will become the single most critical governance challenge for IT departments, as we move from humans clicking buttons to autonomous scripts running with the full authority of a corporate identity. We will likely see a surge in specialized security and governance vendors who focus specifically on these “permission hooks” to prevent agents from escaping their folders or leaking tokens. However, the organizations that will thrive are those that realize there will never be a “ready” signal from the headlines; they will be the ones who have spent the last few years narrowing their cone of uncertainty through careful, experiential usage.
