I was at RSA last week—along with thousands of other security practitioners, innovators, and vendors. It was an exciting and eventful week for us at Concentric AI: we had strong conversations with customers and prospects and were thrilled to be recognized with an SC Award for the Best Data Security Solution, among other highlights.
One of the prominent themes at the conference was agentic AI. We’ve all seen how AI has revolutionized industries and applications, but I found myself asking a simple question: What exactly is agentic AI?
After numerous conversations with fellow AI experts, engineering leaders, software architects, security professionals, and industry veterans, one thing became clear—there’s no shared understanding of what “agentic” even means. What makes an AI system “agentic”? Depends on who you ask.
A big part of the confusion, I believe, stems from the AI community itself. As computer scientists, we have a long history of coining ambiguous or misleading terms. “Agentic AI” joins a growing list:
- Computer science (neither about computers nor a science in the empirical sense)
- Artificial intelligence (a phrase with terms no expert can define precisely)
- Natural language processing (though language is arguably a human construct)
- Data science (not a science in the traditional sense)
- Big data (just… an odd label), and so on
Curious, I asked various AI tools—Gemini, Claude, ChatGPT—for their take on agentic AI. Here’s what Gemini had to say:
If you read closely, you’ll notice that the features attributed to agentic AI—autonomy, planning, adaptability—are the same capabilities that were once promised as part of traditional AI. This kind of semantic goalpost-shifting isn’t new. It’s especially prevalent in cybersecurity, where buzzwords often take precedence over clarity and real innovation.
Rather than chasing the latest label, we as vendors have a responsibility to focus on what actually matters: solving real-world problems, building meaningful capabilities, and educating customers on how our solutions can help—not just renaming what they already do.
The hype cycle is nothing new. Our friend Arthur Hedge from Castle Ventures recently marked the birthday of Claude Shannon, the father of Information Theory. Shannon’s foundational work came out in 1948, but just a few years later, even he was warning against the overhype surrounding his own field. (Links: Post, Comment)
Today, I worry that “agentic AI”—at least in how it’s being marketed in security—is on a similar trajectory: more bandwagon than breakthrough.
At Concentric AI, our Semantic Intelligence platform does what some would call “agentic”: it learns from experience, handles complex, nuanced tasks, makes decisions, and can act autonomously to secure data at scale. Whether that’s “agentic” or not is beside the point. What matters is that it works, and it solves real problems for real customers.
Let’s focus less on labels—and more on impact.
