What happens when AI makes employees more productive today but gradually weakens the expertise companies will depend on tomorrow?
In this episode of Tech Talks Daily, I speak with Dr. Margaret Cunningham, VP of Security and AI Strategy and Field CISO at Darktrace, about cognitive tech debt, the growing risk that companies are gaining short-term efficiency from AI while unintentionally weakening critical thinking, technical expertise, problem-solving ability, and human judgment.
Margaret brings a rare combination of experience to this conversation. With a PhD in Applied Experimental Psychology and a career spanning behavioral science, cybersecurity, privacy, human-centered security, and AI strategy, she examines technology adoption through the lens of how people actually think, learn, develop expertise, and make decisions.
She explains cognitive tech debt by comparing it with the technical debt familiar to software teams. Companies can introduce technology quickly and enjoy immediate improvements in speed and output, only to discover weaknesses underneath those gains later. With AI, the debt may accumulate in people. Employees can appear highly productive while outsourcing the difficult cognitive work required to build judgment, recognize patterns, understand failures, and develop genuine expertise.
We discuss emerging evidence that over-reliance on AI is already affecting professional skills. Software engineers may become less capable of diagnosing problems in code they did not create themselves. Medical professionals can lose decision-making capabilities when they become dependent on automated systems. Across knowledge work, deep reading and sustained concentration are increasingly being replaced by summarization, generation, and superficial review.
Margaret describes the current period as the "bridge years," when AI systems are becoming increasingly capable but people still need to maintain the expertise required to recognize mistakes, question recommendations, recover from failures, and understand when automation should not be trusted. Companies cannot safely abandon human skills before technology can reliably perform those responsibilities without supervision.
The conversation also challenges one of the most repeated promises surrounding enterprise AI adoption: that automation will remove routine work and allow employees to concentrate on higher-value activities. Margaret argues that companies have done a poor job of defining which tasks people genuinely want to give up and which skills they need to preserve. Some of the repetitive, slow, and difficult work being automated may be exactly where people develop pattern recognition, creativity, and professional judgment.
This creates a serious challenge for cybersecurity teams and other high-stakes professions. If employees become reviewers of AI-generated outputs rather than practitioners developing expertise through experience, where will the next generation of senior engineers, security analysts, doctors, researchers, and technical specialists come from?
Margaret explains why leaders need to understand which AI techniques are being used for different business problems rather than treating every form of artificial intelligence as interchangeable. Large language models, machine learning systems, behavioral analytics, and other technologies have different strengths and limitations. Knowing what questions to ask requires domain expertise, creating a difficult paradox for companies that may be automating away the very experience needed to govern these systems responsibly.
We also examine the human consequences of AI adoption. Technical specialists who enjoy solving difficult problems can lose motivation when meaningful work is replaced by reviewing machine-generated outputs. Companies may struggle to understand who owns decisions made through collaboration between humans and AI, while younger employees could lose access to the experiences that previously helped people progress from beginners to experts.
Margaret offers practical advice for business and technology leaders deciding how quickly to introduce AI across their workforce. Companies can identify the skills they need to preserve, create opportunities for employees to practice difficult cognitive work, use simulations and training to maintain expertise, ask teams which aspects of their jobs give them purpose, and resist pressure to automate every task simply because the technology exists.
The message is not anti-AI. Margaret sees enormous potential for artificial intelligence in scientific research, cybersecurity, productivity, and solving difficult problems. But realizing those benefits requires a more intentional relationship between people and machines.
For business leaders, CISOs, technology teams, AI practitioners, and anyone concerned about the future of human expertise, this conversation provides a practical framework for recognizing cognitive tech debt, deciding what should and should not be automated, preserving critical thinking skills, and building healthier forms of human-AI collaboration.
AI can make people faster. The bigger question is whether companies can capture those productivity gains without losing the human capabilities they will need when the technology gets something wrong.
Useful Links
[00:00:00] - [Speaker 0]
It's not your fault that your agentic AI systems are acting outside of compliance. There were just too many datasets to guardrail them all. But with Denodo, you can now organize your hundreds of data sources into one layer and govern your agents with a single approach. Try it now with Denodo by visiting denodo.com to learn more. Are we making ourselves smarter with AI?
[00:00:33] - [Speaker 0]
Or are we quietly handing over the very skills that made us valuable in the first place? And by that, I mean including our critical thinking skills. Because as AI becomes a part of almost every workflow, my guests today will warn that there is a hidden cost that very few people are talking about right now. Her name's doctor Margaret Cunningham and she's from a company called Darktrace. We're gonna talk about the growing problem of cognitive tech debt and why convenience can come at the expense of our own critical thinking and how businesses can embrace AI without allowing all of our human expertise to just fade away and exit stage left.
[00:01:18] - [Speaker 0]
It's a really interesting conversation, this one. But enough from me. Let me introduce you to my guest right now. So thank you for joining me on the podcast today. Can you tell everyone listening a little about who you are and what you do?
[00:01:34] - [Speaker 1]
Yeah. Thanks so much for having me today. My name is Margaret Cunningham. I am a VP of security and AI strategy at Darktrace. I spend almost all of my time working with people within our company and outside of our company noodling through all of the latest and greatest updates in the field and kind of digging into some of the hard problems that we're facing, and doing my best to predict what might happen next.
[00:02:04] - [Speaker 0]
Fantastic. Well, it's a pleasure to have you join me today. One of the reasons you set off my tech spidey sensors was when I heard that you were talking about a concept called cognitive tech debt, and it's a phrase that immediately just grabs attention. But what exactly is cognitive tech debt, and and why do you believe organizations should be paying much more attention to it, possibly now more more than ever?
[00:02:26] - [Speaker 1]
We have known for a long time building technology that we occasionally spend a little bit more effort building things really, really quickly. And we get all of these wonderful capabilities, and it's flashy and fun. And then all of a sudden, we realize that there's not much holding up the building behind it. And then we have to go backwards and try and fix all of those little things that we've let decay a bit. So with the emergence and wide acceptance of different AI services and automations, what we're seeing now is that we're getting the illusion of high performance and productivity.
[00:03:11] - [Speaker 1]
But underneath that, we're outsourcing a lot of the critical thinking and hard work that is really supporting our ability to do this in a resilient way.
[00:03:25] - [Speaker 0]
It's such a powerful point because I think throughout history, we've used technology to reduce physical effort. No problem there. But AI is increasingly, as you said that, reducing cognitive effort as well. It very often feels like we're almost outsourcing our critical thinking, and, obviously, that's a a dangerous thing to do. So, from your viewpoint here, at what point does helpful assistance become harmful dependents, and how do we recognize when that line has been crossed?
[00:03:52] - [Speaker 0]
I mean, looking back before mobiles, we could all remember people's phone numbers, but we don't need to now. But maybe that's that's only a small aspect. But now in AI and so much of what we naturally think about, we just go straight to AI. At what point does it become harmful?
[00:04:07] - [Speaker 1]
So it depends, and I hate that, but I also love that because it's a nuanced topic. I will say recently, Nature, which is a very well established publication, came out with an article called, is AI ruining your skills and the the Outlook doesn't look good or something really ominous.
[00:04:27] - [Speaker 0]
Yeah.
[00:04:27] - [Speaker 1]
And I'm like, wow. And they've consolidated studies showing that surgeons are making worse decisions about diagnoses. Software engineers are losing the ability to understand what went wrong in their code because they're generating things and reviewing versus actually having to do a lot of that hard work. So we're already seeing the impact of those dependencies when we do studies to look specifically for that. This is very risky because, you know, maybe we don't need to remember phone numbers anymore.
[00:05:01] - [Speaker 1]
Maybe we use GPS. Maybe airplanes can fly, and medical devices can kind of be on their own a little bit. But because it's so new, we're seeing gaps in coverage, and we're already having a hard time recovering. So right now, as we're building that dependence, we still have to maintain the skills. We can't just say whatever and offload it very safely.
[00:05:31] - [Speaker 1]
So we're in a I call it the bridge years because we have to do both.
[00:05:37] - [Speaker 0]
It's such an important point because I think many organizations are rushing to automate tasks using AI. But I what are the risks of outsourcing maybe too much thinking to machines, particularly when it comes to developing future expertise, judgment, and problem solving capabilities? All things that we possibly take for granted now, but the next generation coming up, they may not have it if they rely too heavily on AI.
[00:06:02] - [Speaker 1]
So I think we have the illusion that we're still reasoning. We're still establishing expert judgment, which to me is a little bit different from simple decisions. Yeah. But the types of expert judgment that we use in cybersecurity, for instance, are dependent on really understanding and experiencing and growing that type of pattern recognition that we have immediate access to right now. And so deep reading, very rare for people to do.
[00:06:39] - [Speaker 1]
Deep work, very difficult to even find the time to do it. So we're doing a lot of superficial review work that is definitely making it like we have an illusion that we still have that decision making prowess. And it also feels good. It we actually feel like we're doing a lot of really cool stuff, but we're not necessarily building that muscle. So to me, it feels very use it or lose it, and we're not using it very much.
[00:07:12] - [Speaker 0]
Another reason I was excited to get you on the podcast today is because of everything you're talking about here. And your background in behavioral science, I think, gives you somewhat of a unique perspective on human decision making and our habits with technology at the moment. And I've got to ask, for everything you're seeing and hearing out there and observing, are you seeing evidence that our relationship with AI is is already changing how people learn, reason, and approach complex problems? Are you seeing any big changes here?
[00:07:41] - [Speaker 1]
I see a ton of changes, and some of it is simply, do we want to do the type of work that's remaining as AI takes over things? And I have a lot of really fun friends who are technical experts, individual contributors. They love to do the hard work. They love to find a mess and take it out and puzzle with it. And they're not able to do that anymore because they're bombarded with things that they need to review.
[00:08:11] - [Speaker 1]
And so that's very demotivating for people who build a sense of motivation and purpose in their life, in their work, when they're stripped of the types of things they like to do and left with more of a reviewer non creative role. And so I do see that happening quite a bit in, deep experts who are losing a sense of meaning with their work. And, also, of course, the risk of not having a hiring pipeline with expertise and companies not really understanding who did the work anymore. And, I would argue it's very hard to determine the scale of ownership between people and AI. Some things I use it for a lot.
[00:09:04] - [Speaker 1]
Some things I use it for a little. Some days I'm like, let's go for it. But that's really hard, and it's very hard to parse out what you trust the most and whether you think AI was the adequate tool for what you're creating. It's a little messy.
[00:09:24] - [Speaker 0]
In cyber secure, in particular, I think we often talk about attackers exploiting human behavior. And as AI becomes embedded into human judgment and machine recommendations. How do you see all that evolving, especially when we're talking about high stake environments here?
[00:09:42] - [Speaker 1]
In high stake environments, one of the biggest tips that I can tell people is ask a lot of questions about the types of machine learning and AI that are used to provide you a capability. For detection engineering, you might not want to use an LLM, at least not a generative one. For other different types of signal reduction or summarization, sure, maybe. But you need to be much more specific about what the machine is doing for you, the choices that are made in the techniques applied to the problem, and be much more judicious about what you're trusting. The hard part about that is that you have to have expertise even to know the questions to ask.
[00:10:32] - [Speaker 1]
So you have to have the domain expertise of the needs of cybersecurity use cases, threat modeling, defensive postures, what's going on in the threat landscape, and then also be, somewhat fluent in advanced analytic techniques. And this is a very small slice of humans, So we really need to work together to support the types of responsible development, responsible deployment, and responsible use. But there's not necessarily a huge incentive to do that. So
[00:11:11] - [Speaker 0]
And on top of that, I've been to, what, 15 tech tech events this year, and every single one of them, it seems the same phrase is being used that, hey. AI will free your people up from the routine work so they can focus on higher value activities, but our people are important to us. Yada yada yada yada. But, I mean, what is it that separates organizations that are actually using AI to augment human intelligence from from those that might be unintentionally running the risk of weakening it.
[00:11:39] - [Speaker 1]
I think we've done a pretty crummy job of defining what the types of work is that we'd like to be freed from. And on top of that, we've done a really crummy job of identifying the skills that we really need to doggedly preserve. So as we start noticing failures, which I hate to say are going to happen, I think there's going to be a little bit of a clawback on intentional skill preservation, intentional upskilling, intentional practice through modeling and simulation or other other mechanisms, and different types of partitioning what is automated and what's not. And I think that companies that take that seriously, either building products or just large enterprises, will have a much healthier workforce and a much more robust way of managing significant fast tech changes.
[00:12:52] - [Speaker 0]
I would completely agree with you, especially on the crummy job that many are doing there. But my
[00:12:57] - [Speaker 1]
Tons of my crummy jobs. I do. I like to do sometimes I like to do the crummy stuff. Doing some of the crummy stuff helps me slow down. It helps me think about connections that I might not have made before.
[00:13:10] - [Speaker 1]
So getting your hands dirty, doing something slowly, it can spark innovation. And so if we're not doing those things sometimes, we are at risk of losing innovation. AI is average. It's average in many ways. It's averaging across.
[00:13:29] - [Speaker 1]
It's summarizing things. It's taking global stuff and condensing it. That's not innovation.
[00:13:37] - [Speaker 0]
And we do have a great opportunity here. We could have business leaders listening to our conversation thinking, I get it. I've been thinking about this stuff for so long. How can they encourage AI adoption within their organization, but still preserving that deep thinking, skills, their critical analysis, creativity, and expertise that organizations will need more than ever to to survive in that future. What would you say to those people listening?
[00:14:04] - [Speaker 1]
I think a lot about the warnings we get for phishing emails. I know. This is a bizarre segue, but we are always told to not respond when there's urgency, to chill out when somebody is demanding that you do something immediately and quickly. So I I promise AI will be here next month. You don't necessarily have to respond to the urgency of AI mania.
[00:14:37] - [Speaker 1]
Of course, you're probably already using it, But there are ways to strategically slow down, strategically question, and also talk to the people who work for you in critical roles and learn more about what they want to do, the skills they want to maintain, the things that they're worried about losing out on. Because it's not just an exercise in company culture. It is research into the types of spaces where you're going to get benefit from AI and happy teams and maintaining skill sets versus haphazard adoption, which may decrease, de skill your workforce, and also make them miserable. So we have to bring back a human connection to support what that will look like.
[00:15:39] - [Speaker 0]
And I think the speed of technological change that we've seen in the last five years has been breathtaking. It's hard to imagine that just what? Six years ago, working from home was just for the privileged few rather than the many. Hybrid working was not even a thing, and that's before AI burst onto the scene. But if we continue down the current path of increasingly capable AI assistance, what what does a healthy human AI collaboration look like, and what practical steps should individuals listening be taking today to avoid accumulating their own cognitive tech debt in the the next five years ahead?
[00:16:16] - [Speaker 1]
I think we can learn a lot from understanding health behaviors and preventative health behaviors. You know, with things like fast food or smoking cessation or anything like that where you really do have to make effort and buy in and choose healthy paths for your cognitive health when it comes to AI dependencies, it's going to be a challenge. And I think it's going to impact some people differently than others, just like we see access to healthy foods, exercise, and other behaviors that require grit and motivation. So creating that type of workplace where it can be easier to do the deeper, harder cognitive work will also keep a healthier workforce when it comes to this. Unfortunately, we don't know what the long term impact will be, and everything is getting so much more capable so quickly.
[00:17:20] - [Speaker 1]
I don't know. I I I laugh because I was not worried about any of this five years ago. So I'm assuming that I'm probably worrying about the wrong things now.
[00:17:32] - [Speaker 0]
But overall, are you optimistic about the future? Do you think there is that awareness there, and we will naturally come back and seek that that human touch?
[00:17:41] - [Speaker 1]
I think it's giving people a little bit of a, like, stick in the spokes because they're like Yeah. I don't really love how this feels, and I'm gonna go out and touch grass.
[00:17:54] - [Speaker 0]
Yeah.
[00:17:55] - [Speaker 1]
So there is a bit of that. I think that there's a little bit of reconnection because of such a deep emphasis on technology. I also love it. I think it's got huge potential for helping people, for scientific advancement, for protecting organizations from very difficult threats and challenges. But I do think we're gonna have to be very intentional about reconnecting at the human level.
[00:18:24] - [Speaker 0]
I think that is a powerful moment to end on. And for anybody listening that is interested in learning more about finding that right balance between leveraging AI's efficiency and preserving their deep thinking skills, where would you like me to point everyone listening?
[00:18:38] - [Speaker 1]
Well, I love talking about this, so feel free to connect with me. I'm on LinkedIn. Easy to find. But Darktrace also has an enormous amount of materials, including tons of threat research and lots of insights into how we're adopting AI, and what's going on in the AI threat landscape for cybersecurity lovers. So I would definitely check out the work that's happening there, and, I hope to continue the conversation.
[00:19:07] - [Speaker 1]
This is gonna be going on forever now.
[00:19:11] - [Speaker 0]
Yeah. It really is. And I've just loved chatting with you about the risks of maybe outsourcing critical thinking, but more than anything, looking at the tools to get that balance back, enhance human intelligence, not replace it, and and also how to maintain and retain our cognitive skills that ultimately make us uniquely human. So I will add links to everything that you mentioned there. I urge anyone listening that has, really set off a few light bulb moments today.
[00:19:36] - [Speaker 0]
I hope they go and check this stuff out. I think so important, and the level of awareness that we're talking about here will be critical. But thank you for bringing this to life today. Really appreciate your time.
[00:19:46] - [Speaker 1]
Yeah. Thank you so much for having the conversation.
[00:19:49] - [Speaker 0]
I think there's no doubt that AI is changing the way we work. But I think today's conversation offered a timely reminder that efficiency should never come at the expense of our curiosity, our judgment, and critical thinking. And I also loved how my guest challenged us all to think a little bit differently about the relationship between people and AI. And by that, I mean not as a replacement, but as a partnership. But that partnership must strengthen both.
[00:20:21] - [Speaker 0]
And I'd love to hear more about where you stand, your experiences, your insights. Are you striking the right balance between automation and human expertise, or are we all running the risk of creating problems that we will only recognize many, many years from now? Love to hear your thoughts. As always, techtalksnetwork.com, 4,000 interviews across eight podcasts there. You can leave me a voice message, send me a DM, connect with me on socials, or even work with me.
[00:20:51] - [Speaker 0]
Whatever it is, let me know, and I'll get straight back to you. Other than that, I will be back again real soon with another guest. So stay tuned, and I'll speak with you then. Bye for now.