AI Expert Interview: The Benefits And Drawbacks Of Agentic AI

AI agents promise to revolutionize how we work—but what can they really do for you, and what should you watch out for? We asked two of our in-house AI experts who build these systems daily to break down the real benefits, drawbacks, and risks.

According to Barrie Hadfield and Wic Verhoef of Mindset AI, these smart systems can take over complex operations, scale human expertise, and free up your time for creative work. But they also talked about potential pitfalls: unpredictable behavior, security concerns, and the risk of widening technology gaps.

Their insights offer a clear-eyed view of where AI agents excel, where they fall short, and what risks we need to consider as this technology becomes more prevalent.

‍

‍

About Barrie Hadfield

CEO & Co-Founder, Mindset AI

Barrie is an entrepreneur with 35 years of experience building software companies that lead their markets. As CEO of Mindset AI, he’s shaping the future of AI-human collaboration. He believes that AI is here to augment, not replace us.

Barrie has co-founded several successful companies, including Workshare, a document collaboration platform with over 2 million users worldwide. He was CTO for seven years, leading product strategy and development. In 2019, Workshare was acquired by HG Capital. He previously built SkyDox, which evolved into Workshare’s cloud platform after securing £20 million in growth equity funding.

A product-driven leader, Barrie believes great software comes from passionate teams who are empowered and respected. His track record proves it—creating businesses that solve real problems, scale globally, and stand the test of time.

‍

‍

About Wicus Verhoef

CDO & Co-Founder

Wic is a data-driven technologist with a deep passion for machine learning, adaptive systems, and the intersection of humans and AI. As Chief Data Officer and co-founder of Mindset AI, he leads the development of hyper-personalized, AI-driven learning experiences. His expertise lies in making software that adapts to users—rather than forcing users to adapt to software.

Before Mindset AI, Wic was the Lead Data Scientist at Workshare, where he developed machine learning models and decentralized machine learning algorithms. Wic has held analytics and data science leadership roles across fintech, gaming, and consulting. With a Master's in Business Mathematics and Informatics, Wic has specialized in AI, data mining, and statistical modeling.

His work is grounded in the belief that data is the most valuable commodity of the 21st century—and that the future belongs to us, humans who can turn information into intelligent action.

‍

Let’s dive into the interview.

‍

‍

What are the benefits of agentic AI?

Interviewer: Barrie, from your perspective, what are the key benefits of agentic AI?

Barrie: Well, I think the most important thing to understand is that agentic AI is a fundamental shift from how we’ve traditionally approached automation. We've been doing business process automation for 30 years, but those systems were "dumb." They followed predefined sequences, were predictable, and lacked the kind of nuance that you get with human logic.

Agentic architecture uses language models, which embody human logic, and this allows systems to behave more human-likely. Instead of blindly following a set of instructions, an agentic system can actually make decisions based on its understanding of the situation.

Another huge benefit is the scale at which these systems can operate. We’re talking about running incredibly complicated systems like hospitals, ambulance services, or supply chains, which are way too complex for any human or group of humans to handle. Thousands of agentic processes can work together to manage these things. The goal isn't to replace people but to free us from mundane tasks so that we can focus on creativity and productivity.

‍

Interviewer: Wic, what would you add about the benefits of agentic AI?

Wic: I see agentic AI as a way to scale human expertise. For example, in the past, only the wealthy could afford private tutors, but now we have the school system, so mostly everyone can access education. With agentic AI, we can package an expert’s knowledge, methods, and resources into an automated system available to anyone, anywhere.

Another key benefit is adaptability. An agentic system can adapt to a user's needs by using their language, background, and learning styles. It’s about making knowledge more accessible and digestible for everyone, whether that’s through video, audio, or text. It can also adapt to the user's local context and use relevant examples.

From an operational point of view, these systems are incredibly resource-efficient. They can deliver training or education at scale using digital means, making them ideal for diverse settings. Another amazing thing is that these systems can track progress and performance. The data we get from this can be used to iterate and improve the system rapidly.

Finally, we can use these systems to automate boring or repetitive tasks for humans, freeing them to do more meaningful work. For example, we’ve created systems that do this very thing in the coding space. I, myself, use such AI agents, and I am now 3-4 times more efficient than before agents—I managed to remove much of the ‘BS jobs’ from my workload.

‍

In summary, the benefits of agentic AI are:

• Emulation of human logic: Agentic systems use language models to behave more like humans, allowing for decision-making based on the situation rather than rigid processes.
• Dynamic decision-making for complex systems: These systems can adapt to complex situations with more nuance and creativity. Agentic AI can manage hugely complex systems, like hospitals and supply chains.
• Augmentation of human capabilities: It's about freeing up humans to do more creative and productive work by taking on more mundane tasks.
• Scaling human expertise: We can package expert knowledge into automated systems, making it accessible to a broader audience. These systems can adapt to user needs, personalizing content to language, context, and learning styles.
• Resource efficiency: Agentic AI provides cost-effective solutions that can be applied in various contexts. They also offer valuable data for rapid iteration and system improvement.

‍

‍

What are the drawbacks of agentic Ai?

Interviewer: We've discussed the benefits, but what are agentic AI's potential drawbacks or limitations?

Barrie: One of the main things to understand is that agentic AI is not a magic wand. You can’t just throw a few workflow steps together and expect it to work perfectly like a business process automation system of the past. It's a mistake to assume you can just "hope for the best." Essentially, a language model is a new form of logic engine that embodies human logic, and that logic is inherently unpredictable.

The fact that it embodies human logic is both a benefit and a drawback. Human logic is chaotic and creative, which might not always be what you need in a business process automation system. A traditional system is predictable and rational, whereas an agentic system could behave irrationally. It's about recognizing that you are introducing a less predictable element, and you need to have robust testing systems in place to ensure that it behaves the way you want it to. If a system suddenly starts ordering a thousand cans of tomatoes for a restaurant, you need to know why and how to correct it.

Also, it's important to recognize that if these systems are trained on human behavior, they will make human mistakes. It's an assumption to think they will automatically learn from those mistakes and retrain themselves. That sort of self-learning capability has to be built into the system. Ultimately, agentic AI is not like machine learning, which is about optimization and getting better from many examples. Agentic AI uses logic, which is unpredictable at its core. We have a massive responsibility to bring this technology to practice with caution.

‍

Interviewer: Wic, what are your thoughts on the drawbacks of agentic AI?

Wic: One of the biggest hurdles is building trust. It takes a big leap of faith for a business to trust an agent to act as an ambassador to their users and customers. We need rigorous testing and evaluation frameworks to build this trust, so we’re working on a robust evaluation system that can give us and our clients confidence.

There are two parts to an agent: the infrastructure it runs on and the agent configuration. The key is that we, as developers, must be sure that any changes we make to the infrastructure are improvements and that they haven't broken anything. Equally, our clients need a way to understand the impact of the changes they make in agent configuration. A robust testing framework is invaluable for both these.

There are certainly risks around security and reliability. We need to ensure that data is kept safe and IP is protected. It's essential to keep different clients or users of an agent separate so that no sensitive information is leaked.

Another thing to consider is that these systems can learn from user behavior and feedback, but we need to ensure that this feedback comes from trusted users. We can't assume all feedback is equally valid. Also, whilst we aim to make the systems mimic the behaviors of the users they represent, we want to be sure to scrub any intellectual property.

There is also the possibility that agentic AI could widen the gap between those who have access to and understand the technology and those who do not. It's still to be seen if it will improve humanity's collective intelligence level—or not.

‍

In summary, the drawbacks of agentic AI are:

• Not a "magic wand": Agentic AI is not a simple, one-size-fits-all solution and requires careful implementation. It's not a case of "hoping for the best."
• Unpredictability: Human logic makes these systems less predictable and potentially irrational, which is not always ideal for business process automation. They can reproduce human errors.
• Need for robust testing: Thorough testing systems are essential to make sure the systems are working as expected and for remediation.
• Trust issues: Building trust in AI agents is a big challenge.
• Security and reliability: It's essential to keep data safe and IP protected and to prevent data leaks between users or clients.
• Need for trusted feedback: Feedback needs to come from trusted users to avoid skewing the system.
• Potential for increased inequality: These systems may increase the gap between those with access and those without access.

‍

‍

What are the risks of agentic AI?

Interviewer: Now, what specific risks or ethical considerations are associated with agentic AI?

Barrie: Agentic AI can be used for any purpose, good or bad, as it’s essentially a tool for humanity to achieve its objectives. So if someone has a criminal objective, or a political objective to manipulate people, they can use these tools to do so. There’s no way of preventing this, just as a telephone can be used to call loved ones and make scam calls.

I think that we have to accept that bad actors will use AI to manipulate the world, as we see today with online media platforms. Regulation is needed, as current laws are insufficient. For example, current media laws mean that online media platforms cannot be held accountable for the content that is on them, whereas magazines and newspapers are. This is something governments and large organizations should be looking at. If we leave it like the "Wild West", that’s what we’ll have.

Wic: I agree that the potential for misuse is a significant risk. As with any powerful technology, agentic AI can be used for ill gain, just as a bad teacher can do harm.

Security and reliability are also concerns. We need to ensure that the data people upload into the system is kept safe and that intellectual property is protected. A data leak could be catastrophic for any company. It’s also important to ensure that information is not leaked between users or clients.

Finally, we are still in the early stages of development of these systems, and they are evolving rapidly. This means there is an ongoing risk that the system may become outdated or that some better technology comes along. It's essential that our systems are resilient against tool changes, so we can adapt to new technology as it becomes available. This means being able to swap out language models, vector stores, and other services with minimal effort.

‍

In summary, the risks of agentic AI are:

• Potential for misuse: Agentic AI can be used for harmful purposes, including criminal and manipulative activities.
• Lack of accountability: Current regulations are not sufficient to hold online platforms accountable for the content on them, which can be exacerbated by AI.
• Security and reliability: Data leaks or the loss of intellectual property is a very real risk.
• Technology obsolescence: Rapid technological advancements in AI mean systems can easily become outdated.

‍

‍

Will AI agents replace humans?

Interviewer: A prominent discussion is centered around whether AI agents will replace humans and how the workplace of the future will look.

Barrie: I feel very strongly that AI is not going to replace us. I see AI as an extension of human evolution, rather than a replacement for it. In the future, I think we will have the opportunity to have more time in our day, to be more creative, and to be more productive. AI will take away a lot of the work that we don’t want to do or the complexity and annoyances that we have in our lives. It will also help us to make more objective decisions, rather than subjective ones.

In the workplace of the future, I think that the human role will be to tell AI systems what to do and to oversee them, to make sure they are working properly. It will be a transformation across the board. Just like in banking where computers have completely removed the need for people to stamp paper—there are still many people working in banking, just with different jobs.

Wic: I agree that there will be changes in the workplace, and I agree with Barrie that AI will not replace humans. There will certainly be impacts.

I know from my own experience that AI tools can make me, as a developer, much more productive. There are many "BS jobs" that people do right now, and if this technology can liberate people from those tasks, then that is a good thing. It's unfortunate if it takes away the livelihood of some people—I hope the technology we produce will also allow people to upskill themselves. Ultimately, I believe that AI will shift us from being creators to being editors, and that's awesome.

In the future, it is likely that more people will be able to get more done and focus on more meaningful work. It is also important to recognize that not everyone will benefit equally, and there will be a gap between those who have access to the technology and those who don’t.

‍

‍

Exciting future developments in agentic AI

Interviewer: Wic, I know you’ve been busy working on some exciting developments related to the topics we covered. Can you share a bit about these projects?

Wic: Yes! There are a few things we're working on right now at Mindset AI that I'm really excited about. One is our new approach to RAG (Retrieval Augmented Generation), in which we are adapting an agentic form of  RAG. In this there is a series of sub-agents under the hood which decides on the best interpretation of the user’s intent and then work together on the best approach to retrieve the best data to answer the user’s question.

The result is that we can now get a much more nuanced answer to the user's query, and the system can avoid doing unnecessary searches. This multi-agent approach allows us to address users’ questions with more accuracy.

We are also improving the way feedback is used so that the system can immediately learn from trusted users. If you, as an admin, give explicit feedback to the agent (by using the  like or dislike buttons) or implicitly hints that a specific agent response was good or bad, the system will learn from this feedback and improve future responses. It's a very rapid loop, and the agent starts to mimic you, and this can create very specific agentic ambassadors for our clients.

On a related note, we are also working to ensure that our system is resilient against tool changes because the market is moving so fast. New LLMs and other services are appearing all the time, and we need to be able to adapt. So, we are working to ensure that we can quickly swap out things like LLM for another with minimal disruption to our users.

Ultimately, our goal is to iterate as fast as possible and provide our clients and their users with a reliable AI system that just works.

‍

‍

Conclusion

While AI agents aren't magic bullets, they're opening up exciting possibilities for how we work. They can handle complex tasks and free us from the boring stuff, even if they sometimes act in unpredictable ways (they get their logic from us humans, after all!). The message from our experts is clear: AI won't replace us, but it will help us spend more time on work that actually matters.