The conversation around AI is changing quickly. That’s why I’m so excited to share our podcast series Ready. Or Not. We’ve paired comedian Nathan Macintosh with expert guests to talk about AI agents, cyber resilience, trust, data management, and more.
In our first episode, Nathan sits down with Dr. Reid Blackman, founder and CEO of Virtue Consultants, to tackle one of the biggest topics in AI today: agentic AI. From ethical challenges to security risks, their conversation explores what happens when AI moves beyond making content to making decisions – and taking action.
One thing is clear: Agentic AI isn’t just another technology trend. It’s changing how we think about decision-making and the role AI will play in our organizations. If you’re wondering what agentic AI means for your business, this podcast is a great place to start.
Watch the full episode on Readiverse.
Blog Key Takeaways
- Most AI failures are caused by unintended consequences, not malicious intent.
- Agentic AI can access systems, tools, and data to get work done, making it both incredibly useful and inherently risky.
- AI agents can create new security challenges, from prompt attacks to expanded attack surfaces.
- Multi-agent systems can increase efficiency, but they can also amplify mistakes when systems are connected.
- Organizations need practical frameworks for managing AI risks before they become real-world problems.
First, Do No Harm
One takeaway from the episode is that most AI failures don’t start with bad intentions. Many begin with organizations trying to solve legitimate business problems.
Dr. Blackman uses a failed Amazon AI recruiting tool as an example. The algorithm was trained on past resumes and hiring data to inform future hiring decisions. AI ultimately learned patterns that favored male candidates because those patterns existed in the data.
The result wasn’t what Amazon intended, but that’s exactly the point. AI systems can learn lessons we never meant to teach them.
What happened next was encouraging: Amazon tested the system, identified the issue, tried to correct it, and ultimately discontinued the project when the problem couldn’t be resolved.
We tend to treat AI failures as proof that technology can’t be trusted, but Dr. Blackman makes a different point. Responsible AI isn’t about pretending mistakes won’t happen. It’s about testing, learning, and being willing to stop when something isn’t working the way you intended.
When AI Becomes Your Coworker
The Amazon example also highlights something bigger. AI is capable of delivering value, but it can also produce unintended outcomes when we don’t fully understand how it’s learning or making decisions. Generative AI showed us what AI can create. Agentic AI is showing us what AI can actually do when it’s connected to business systems.
One comparison that stood out to me was that agentic systems are, in some ways, starting to look like employees. To be useful, they need access to the same tools, databases, and software that people use. Give an AI agent access to one system, and it can do one job. Give it access to dozens of systems, and it becomes more powerful.
“More access means more capability, but it also increases risk dramatically.”
– Dr. Reid Blackman
Sneak Peek: Keeping AI in Check
What happens when your AI agent starts interacting with other people’s agents? In this clip, Dr. Blackman explains why monitoring multi-agent systems will become one of our biggest challenges.
A New Kind of Security Challenge
Agentic AI changes more than the way work gets done. It also changes the way we think about security. Instead of following predefined workflows, users interact with AI through natural language. That makes these systems more intuitive – but it also creates new challenges that traditional software doesn’t have.
As Dr. Blackman explained, attackers don’t necessarily have to break into an AI system the way they might traditional software. Instead, they may try to manipulate it through carefully crafted prompts that influence its behavior, bypass safeguards, or expose information it shouldn’t access. It’s a reminder that as AI becomes more capable, security has to evolve right alongside it.
“Do we need AI watching AI?”
– Nathan Macintosh
The Risks of Multi-Agent Systems
If one AI agent can make a mistake, imagine what happens when multiple AI agents start working together. While it may be more efficient for systems to be connected, it also creates more opportunities for failure.
If one agent makes a mistake, it can create a ripple effect. A small issue can become a much larger one if organizations don’t understand how those interactions work. This doesn’t mean multi-agent systems are inherently risky. It simply means they require the same level of planning and oversight that organizations would apply to any complex business process.
“I learned about agentic AI today and I’m already scared. Now you’re telling me AI agents talk to other AI agents?”
– Nathan Macintosh
Do You Know Who Your AI is Talking To?
If managing your own AI agents sounds challenging, consider what happens when they start interacting with someone else’s AI. You may know your own guardrails and policies, but external AI systems may be different. You may not know how they were trained, what they can access, and if they have the same safeguards in place.
Get Ready
Agentic AI is moving quickly, and the technology will keep evolving. The organizations that succeed may not necessarily be the ones that adopt AI first. They’ll be the ones that understand how to govern it, test it, and build trust around it.
One of the goals of Ready. Or Not. is to move beyond the hype and examine what responsible technology adoption actually looks like.
Dr. Blackman’s perspective is a reminder that successful AI adoption isn’t about choosing between innovation and caution. It’s about balancing both. That’s exactly the kind of conversation we’re excited to continue throughout our series.
Watch the full episode on Readiverse.
FAQs
Q: What is agentic AI?
A: Agentic AI refers to AI systems that can take actions, access tools, interact with applications, and complete multi-step tasks with varying levels of autonomy. Rather than simply generating responses, they can actively perform work across connected systems.
Q: Why does agentic AI introduce new risks?
A: Agentic AI often requires access to multiple systems, applications, and data sources. While that access increases usefulness, it can also expand the potential impact of mistakes, misuse, or security compromises.
Q: What are prompt attacks?
A: Prompt attacks involve using carefully crafted inputs to manipulate an AI system’s behavior, bypass safeguards, or expose information that should remain protected.
Q: Why is monitoring becoming more important?
A: As AI agents become more autonomous and connect to more systems, they are also becoming less predictable. Monitoring helps organizations identify unexpected behavior early and understand how AI systems are interacting with people, data, and other AI agents.
Q: What are multi-agent systems?
A: Multi-agent systems consist of multiple AI agents communicating and collaborating with one another to complete tasks. While they can improve efficiency, they can also introduce additional complexity that organizations must manage carefully.
Q: What’s the biggest takeaway from this episode?
A: AI risk isn’t just about what technology can do. It’s about understanding how systems behave when they interact with people, data, applications, and each other – and putting the right safeguards in place before problems arise.
Katherine Demacopoulos is Senior Director of Global Content Strategy and Programs at Commvault.
