Swapnil Bhartiya: Hi, this is your Swapnil Bhartiya and we are here at Open Source Summit in Minneapolis and we have with us once again Clyde Seepersad, SVP and GM of LF Education. First of all, it’s great to have you back on the show.

Clyde Seepersad: Thanks for having me back.

Swapnil Bhartiya: The general thinking is that AI is killing jobs. And that’s when you hear layoffs, you feel that, you know, they say because of AI, but I don’t think that’s entirely true. And you have some data that actually shows the opposite side of the picture. Based on your finding, it feels like the real challenge is readiness, security, maturity, skill and people. So there’s so much to talk about today which may change people’s perception. Before I go deeper into all the statin number, let’s just ask flatly, is AI really killing jobs or not?

Clyde Seepersad: The answer is in general, you don’t believe what you read in the headlines. What you are reading in the headlines is a slice of reality which is for the very largest employers. And we did this in the survey, we categorized it as 20,000 employees or above. And really it’s 100,000 employees or above. So when people talk about FAANG, you know, Facebook, Google, the very largest employers have cut over 100,000 jobs. And that is true. And of course any of us who spend any time reading the headlines, you think, oh my gosh, this is the apocalypse. Everybody is shedding jobs. When you go to the survey data and you speak more broadly, they are actually the exception smaller and medium sized organizations. So below 1,000 and between 1,000 and 20,000, those organizations have all been adding technical jobs and they’ve been adding them faster than the big companies have been shedding them. And so yes, it’s true that a lot of people have been displaced and that they got laid off of the very big tech companies. What’s also true is most of those people have found new work within smaller tech companies relatively quickly, so that the net impact for technical audience hiring is an increase. Then there’s a second part of the story which is there are a few specific places where AI has reduced employment. And those are the places that you would think of as sort of the low hanging fruit. So contact centers. I used to run a contact center in Manila. A lot of what were called tier one employees have been reduced because what would they do? I lost my password, I don’t remember the URL. How do I retrieve my certificate? Well, it is true that there are a lot of what Tier one employees and I say Manila, but it’s actually Everywhere where we don’t need people to answer those questions, we need people for the more complex inquiries. So yes, there are a few places I think the contact center business is easily the primary case. The jobs that were affected in those cases aren’t the technical jobs. They are the jobs that the agentec technologies are good enough to be able to start supplementing. So you have two real data points. Contact centers, very large technical employers where employment has been reduced. Nobody is writing a screaming headline saying mid sized business hires 20 technologists. That does not feel like news. But that actually is what’s happening and it’s happening at scale. It continues to be true that so long as you’ve kept up to speed and we’ll talk about breadth in a second and the importance of a portfolio of skills. If you keep your portfolio of skills, there’s actually more opportunity than there was before the agentic area because the appetite to do things agentically has gone up significantly. But it is hard for people because I understand the disconnect between the screaming headlines of 5,000 jobs here, 10,000 jobs here, and a quieter reality that in aggregate we’re bringing more people into technical careers.

Swapnil Bhartiya: Small and medium sized businesses, they actually make up almost 50% of the economy. So they are actually the backbone. The big ones are backbone because of the revenue, but they run the country. So talk a bit about why is it that big organizations, they’re laying off, but smaller, they are hiring.

Clyde Seepersad: Part of this is the context of what happened after the pandemic, which is every tech company shot, stock went through the roof, supply was vastly short of demand and everybody was just hiring frantically. And if you’re a big tech company and you’ve got the stock options and you’ve got the ability to pay salary, they were a disproportionate draw of technical talent into the largest technical employers. Those are the technical employers that are now said, we probably overdid it. So we’re going to let some of these folks back in. The small organizations had been starved of talent. So there’s two things happening, right? One, they had been starved of talent because they literally couldn’t afford to keep up with, you know, Faang Co. Made me an offer. And the second is that now especially that people are beginning to get a sense that agents can do things that are really powerful. The ability to move the needle with agentic workflows is ironically higher the smaller your org is because the bigger and more complex your org, 25 different stakeholders and four sets of lawyers and multiple jurisdictions to think about, right? It gets quite challenging. You’re a small manufacturer here in the Midwest, you have your thing, you know what you can do to improve your supply chain, you know what you could do to improve your communication flow with your clients. And you can do that. You bring a couple of technical folks in, you help them figure out how to unlock the value from your legacy tech, and bang, you’re off to the races, right? And so the appetite is bigger, the access to talent is bigger, and that’s why they’re making those investments.

Swapnil Bhartiya: What are some of the major concerns these days? Ironically, 99.9% interviews, discussion that I have, irrespective of which industry I’m talking about, is all about Agent tki And few things keep popping up in my discussion. Governance, security. That keeps popping up because yes, now agents are capable of taking actions. They are not just giving you a feedback. They can go and execute things without human in loop. So they do have to put. So what are you seeing there? What are the kind of concern organizations have?

Clyde Seepersad: You know, I think the elephant in the room is security. And if people weren’t aware of it before the whole drama around the limited release of Mythos from Anthropic and Project Glasswing and just how insanely powerful this new generation of frontier model is at not just finding vulnerabilities, but stacking the vulnerabilities. Right? So I was using the analogy of you have your jewels in the safe. What the agentic models are able to do is figure out, okay, first there was a hole in the fence. Well, the hole in the fence doesn’t seem like that bad. And then there was a broken window to the basement, and that in itself doesn’t seem that bad. And then there was a way to unlock the door to the bedroom, and that in itself doesn’t seem that bad. And then there was a way to tell where the key for the safe was. When you could string all five of those things together, you now have a major problem, right? So what seemed like maybe five independent non critical vulnerabilities about been stacked together into one massive vulnerability. And that’s what we’re seeing, right? And so there’s this huge wave of apprehension about, wow, there’s a lot of cyber risk footprint. And that’s before we even talk about all the people vibe coding on the business side that have no clue about how to think about, is it okay if I give root access to this directory, which is what the agent prompted me to give? And I was like, sure, if it makes it work, I’ll give it the root access. So there’s all that surface area, right? A fantastic increase in the surface area, the attack surface area on the security side. And then you couple that with this demand of well, I want to move more stuff into the technical space. I want to take things where I was emailing invoices back and forth and corrections and I want to automate that. I want to automate more on the, you know, the fulfillment side. So there’s more attack surface just within the software and now there’s more demand for doing more stuff with agents. And then the third thing is we have not yet begun to have a modern infrastructure within which agents operate. So today in the keynote, Jim was talking about X402 for payments between agents. Really nothing exists. Today we announced the intent to form the ans, the agentic version of DNS. How do we know an agent is real and what’s the provenance behind the agent? And there are much more things to come, right? So for instance, even if I know the provenance of an agent, how do I what is required for me to trust allowing that agent to execute? Because in the real world, swap does something bad to me. His reputation is ruined, I will never do business again in the agentic world. The agent does me wrong, somebody kills the agent. There’s no accountability. So maybe there’s another layer of how do agents put stakes up, right? It’s like, hey, I want to do this transaction and I’ll stake you $5,000 in case I do something horribly wrong. Yeah, that’s fine. That’s about their level of risk. So what’s the insurance In a world where trust and verifiability and reputation doesn’t matter because it actually doesn’t matter because the agents, you can shoot the agent and spin a new one up tomorrow, get a new ANS listing for it. So there’s a lot of unresolved questions about as we move from baby steps into full scale production, what are all the pieces of the Agentix stack that do not yet exist because we haven’t gotten that far, that have to do with security, trust exposure, liabilities, the whole nine units.

Swapnil Bhartiya: I will go back to the data that you mentioned. What I hear is that of course hiring is increasing almost 31%. That as you. But a lot of companies are also reporting almost 57%. Companies still report a security capacity gap. How do these two things exist? Because hiring should actually take care of security.

Clyde Seepersad: Well, so I think one of the fundamental things we’re realizing, and this is an old realization, but it has a New urgency. In the age of AI and agentic, security is a skill and a function, not a job role, right? So we had started to get comfortable. Organizations had a CSO and a security team and there was this idea that you could focus on speed, get it out, get it out, get it out release cycles and then the security folks would come clean up the mess, right? And they would figure out, oh, maybe tighten this here. I think we should all be totally over that fantasy of security is somebody else’s job. Security is all of our jobs, right? One of the things we did at the Linux foundation last year actually in 2025, start of the year, is we launched this thing called the cybersecurity framework, which was a way to contextualize specifically at the job family level, the security responsibilities of people whose day job is not security. I’m a database administrator. What are my security responsibilities? I’m a network administrator, I’m a front end developer, I’m an app developer. I think we had 14 job families because that was an attempt to say security is a skill and a function embedded in every role in technology. Not a specialist or not solely a specialist function that comes in and gets layered on top like a layer of icing. The urgency of that is even greater now than it was in January 2025 when we launched the cybersecurity framework. Because the awareness of just how much footprint of risk there is and that has implications for how we hire, that has applications critically for how we upskill the people we already have working on these things to say, hey, you know, we used to not press on talking about the security protocols you’re taking, but now we’re going to have to insist that you talk about it in the headline, not the sub footer.

Swapnil Bhartiya: No. And that also brings a very interesting point. I think I had a discussion with you. I mean I think we had this. It’s much cheaper to train your, you know, than to hire. So talk about upskilling. Your existing team shows nearly an eight fold advantage over external hiring for business. So why do half of organizations still default to recruiting outside?

Clyde Seepersad: So we’re old enough to remember, and you don’t actually have to be that old to remember the great pursuit of cloud native talent. Seven, eight years ago, everybody wanted to hire people with container skills and containers in production and kubernetes orchestration. And there was this mad rush, right, to just poach. Everybody was poaching everybody else. One of the things that feels so different now is people are almost embarrassed to raise their hand and say I want to hire a new employee, it’s gotten to be uncool to say I want to increase my head count. Imagine that. Increase my head count. So part of it is there’s this weird cultural stigma all of a sudden with I’m hiring new humans, which came out of this mix of AI has all this power and then agentic has all this power. I think that then coupled with the fact that as people began to experiment particularly with the agentic workflows, they’re realizing, wow, the agent can’t get to my database because my database thinks all these six requests a second are a DDoS attack and shuts it down. Well, the agents move at that speed and so I can’t even get into my infrastructure. So now there’s this rolling wave of realization that to get to this promised land of AI magic and theories, you have to redo huge bits of your infrastructure to even make it compatible, to even make it visible to an agentic workflow in the first place. How am I going to do that? It’s going to have to be Bill down the hall that knows who’s the only one that knows how to open up that service and to begin to recompile it into something. Right. And so now Bill has to be upscale because they’re going to have to move this stuff to be a cloud native workflow to make an API available on it. So I think it’s both things, right? I think it’s this idea of, well, if it’s not cool to hire people, maybe I should have my existing people do it. Plus, wow, most of my infrastructure is not compatible with what it needs to be to run an agentic workflow. Again, I need my existing people to do it. Either way, there’s a whole newfound idea of maybe I should upskill my people, which is great. As the head of education, I think I’m always excited when people decide to double down and invest in their existing staff.

Swapnil Bhartiya: Yeah. And while here we are talking about upskilling your existing workflow, what about entry level jobs? Are they growing? Because what I was hearing is that because of all those agent AI organizations are not hiring entry level because students are still coming out. They’re still going to computer science. What does it mean for them? Parents are also nervous. What does that mean? Because if you don’t get hired as an intern or at the entry level, how do you get five years experience in agentic AI?

Clyde Seepersad: Well, you know, I feel a little bit like this, like I feel about my teenagers where, you know, when kids are young, you go Through a phase where you drop everything in their lap, right? And they want XYZ and you show up and you provide it to them. And I feel like without stretching the analogy too much, we had gotten to a place where in large part our computer science graduates, they would just sit on campus and the recruiters would show up and There would be 10 recruiters and they would be talking and they’d have five job offers and they could nitpick and they could decide and that has gone away, right? The employers are not coming to campus. Everybody is freaking out. But you talk to the kids who got a job on what did they do? First off, they went and learned more than what was in the minimal CS curriculum. They went and taught themselves cloud native. They went and taught themselves DevOps. They did put OpenCloud on an M3 box and tried to fire it up and see what it was all about. They did go try to experiment with agentic workflows. And it’s not to say that they had to make themselves an expert, but they had to have the intellectual curiosity to broaden out and think about the entire technology stack and begin to develop some skills and some familiarity with all of that. Plus they had to go try to embed themselves in something, submit a code patch, get it pulled, go to meetup groups, see who the local employers are, try to make connections. And I think what you’re seeing is this model of how do I find a job is changing from I sit on campus and wait for the job fairs to I have to get out there and network. I have to get out there and volunteer. I have to get out there and find projects I care about, startup companies to talk to and find the founders. And when I do that, I have to be able to show that I’m more than just a dev, that I understand what the whole stack looks like. Because maybe one day they’re using me for figuring out how to rebuild an existing service largely through coding assistance. Maybe the next day I’m helping them think about security implications. Maybe the next day I’m helping them troubleshoot something for the customer front end or build a demo. And so there is this, the responsibility on the young graduates to show that the college credential used to be necessary and sufficient and now it is arguably for sure not sufficient and possibly not even necessary. So they have to do a lot more work. And this is an area where I would really challenge the universities to say, hey, we have got to do a better job of preparing these young people. We’ve got to give them Access and exposure not just to the broader range of technical skills that they need, but also to these ways of being where you’re getting out and you’re engaging with the real world, you’re trying to contribute to projects and you’re trying to get internships and you’re trying to build a personal connection. Because while all of this has been happening, the other thing that makes this extremely difficult if you’re right out of university is the AI slop battles between the AI generated applications where the applicant is basically not even looking because they have an agent writing the applications and then the AI managed hiring screeners which are violently screening out the 50,000 applications you got. Like that is a zero sum dead end battle if you’re going to try to get a job by brute forcing your way through the online web apply indeed portal because the AI wars have come there first. Much better off to have human connection, to have met the founder of a business or the owner of a business or the head of an engineering team, found a way to establish a relationship. So I think this idea of there’s getting to be such a premium on human interaction and Trust and familiarity versus I’m one of 10,000 emails in an all AI generated stack. It’s easy to see how that is not part of the playbook anymore. Your chances of winning that lottery are getting lower by the day because the number of AI generated slop applications. We opened a job, I got 1000 applications in a day. It’s just, it’s insane.

Swapnil Bhartiya: Are you folks working with universities to help them navigate this or that’s trying it is extra.

Clyde Seepersad: I have a lot of empathy for the universities because they have very, very specific requirements to stay accredited. Right. So if you’re not accredited, students can’t get loans, nobody wants to take your program. But to stay accredited you have this insanely long checklist of things you have to do, curriculum you have to cover. And much of that stuff doesn’t reflect modern computing and practice. So they’re stuck in this bind where they have this massive checklist which they have to do to stay accredited. But that checklist, the overlap between that checklist and what employers expect is has been shrinking over time. It has dramatically now mismatched in the age of AI and agentic and cloud native. And the accredited agencies have not caught up. And so we’re still delivering the 1990 version of degrees to the 2026 version of college kids. And it’s a really unfortunate thing and we’ve got to figure out how to better prepare kids as part of the college program, but that that’s a bigger ecosystem challenge, and that’s going to require a whole lot of other people to come together to figure out how do we revamp education, specifically in computer sciences, in an age where the old checklist of what it takes four years to do has so much that is interesting but not really relevant to what you as an employer expect this person to know. Day one, what would make them attractive to you?

Swapnil Bhartiya: Day one, how do you envision the organizations will look like? Because if we have that perspective, then whether it’s college graduate, whether it’s organizations, whether it’s people who have been laid off, they understand this is how the organization look like. This is where I fit in.

Clyde Seepersad: Yeah. I mean, I think that two aspects of that, right. One is there is an enormous amount of knowledge of how a company works that is implicit. It’s the one guy who remembers this thing we tried five years ago that blew up horribly because, you know, some factors that nobody considered. And this other person that remembers, hey, we talked to clients about this way of doing business, and it doesn’t work for their workflows because X, Y and Z and those things don’t show up anywhere. Right. They’re held as implicit knowledge. There is insane value in the implicit knowledge because the agentic workflows are very good at doing the things that are incredibly repetitive, incredibly well defined, incredibly low levels of variability. And once you get into where it’s not highly deterministic, the judgment and experience of all the things you’ve learned over time becomes incredibly valuable. And so I think the last time we talked, we talked about the fast food, right? And the ordering and the fact that it didn’t allow for the fact that people changed their mind after they said, yes, the order is confirmed. And that is in fact what happens some, you know, maybe majority of the time. Right. So you have to have all that context made available. At the same time, it is true that you can never sit. I think we’re at a point, especially in technology, where your victories of yesterday don’t matter for tomorrow. So we all have to constantly be just pushing ourselves to say, okay, what’s the next thing I need to learn? Or what’s the next thing I need to learn? And so the idea that we used to get to sort of the middle of our career and we’ve established this wonderful foundation of deep knowledge, and that was the sort of like, springboard for us for the rest of the career, I think that has become a lot less true. It’s much More what have you done for me? Late nil. And so I think all of us have to get accustomed to this idea that it’s like the shock, right? You gotta keep swimming. You gotta keep swimming. That’s a big change. I think that’s a big change for those of us who are more senior in our careers. This idea that we can’t sit still and that into our 40s, 50s, 60s, we have to keep learning these new technologies, figuring out how to adapt to them. The great irony in all of this hesitation to hire young people who are the most people most hardwired to try the new stuff, the youngest people, right? And so you have people who are on the one end, the experienced people, who are you, who you’re having to coax into this new reality that every day you should be developing new and different skills and you can’t just sit back and rest on your laurels. And on the other hand, you have all the kids who want to come in bursting with ideas about how to do things dramatically differently. But we’re hesitant because it seems like something’s wrong if you’re trying to hire head count. And so, you know, the old playbooks have just disintegrated in our hands. And we’re all figuring out what does the new set of playbooks look like. But we’re also discovering that the playbooks still involve people. The agents don’t sit bolt upright at 2am with a bright idea of, wow, we should think about doing X. They don’t have inspiration in the shower. They’re very good at responding to the things we request them to do. Hey, go look at XYZ and see if we can find a correlation. But they don’t have innate curiosity. They don’t have the aha moments. They don’t have that ability to like, connect these seemingly unrelated fields into, like, inside. Now, that may not be true. If ever we get true AGI, then maybe that’s not true anymore. But for right the second, the impetus of what to ask the agent to do comes from a human. The genesis of why the human is asking that is often born of insight into the business and how it works. Or sometimes from the outsider perspective of the brand new person coming in saying, I can’t believe we even do this. Like, why wouldn’t we blow up the process and start over in this different place? And so there’s value in those perspectives that the agents can never capture because the agents don’t have, ironically, the agency to think and decide what should be important. What should I, you Know the agents don’t wake up and think what should I accomplish today? Right. They’re waiting for the instructions, they’re waiting for the human operator to say, go do xyz.

Swapnil Bhartiya: If C level executive walks to you and you’re like, no, we are seeing things. You are presenting a different kind of picture. Next Monday I have a meeting. How should they prepare? How should they do things differently within their organization to take advantage of both people and AI the most?

Clyde Seepersad: The thing I recommend to people in general if they’re as they’re newer to the process, is ask your peers, compare notes. You know, my most common use case right now is not agentic, it’s model based, which is before meetings. I will ask the agent, I’ll ask the model for our summary of I’m meeting with this person to discuss X, what has been, what’s most likely to be on their mind and what have they been talking about. And I get this amazing summary from a blog post here to a video that they did there, to a posting on LinkedIn here. And it’s incredibly helpful for me to prepare because it helps me better understand, ah, that’s what this person has been thinking about and talking about publicly. Totally an idea I stole from a friend who was telling me about him. You know, he was using it in a context and I just thought, wow, why didn’t I think of that? That is such an incredibly powerful use case. Right. Because it’s literally would have been untold hours of research finding the old YouTube clip or the old podcast appearance, but it’s very, very impactful to. And I think it helps them get more value from the conversation because, because I show up better prepared to engage with them on so, you know, talking to people, looking at practice, trying to figure out what are ways in which other people are doing this. Because I think it is true and these technologies, they’re so new, it’s sometimes hard to just spontaneously think about oh, this is how I’m going to use it if you’re not. Now of course we’re talking about the non technical audience, right? And how do you sort of quote unquote, regular business people sort of get value from these systems. But the other bit you were just talking about and I wanted to come back to because it’s, I think it’s really important. One of the things that happened over the last hundred years is we got a much bigger corpus of knowledge with a much, with massively more silos of specialization. Right. And so you used to be a computer developer. Computer programmer, and then it became that you were a front end, which is a backend developer. And then it became that you were a specialist in front end for web architecture and so on and so forth. And there’s a lot of experience and expertise that built up right over all these different specializations. And I think we feel a tendency now to devalue that because you just go ask the machine guards and it tells you, oh, to do so. And so you blah, blah, blah, blah, blah. I think the thing we need to remember is that is only the illusion of expertise. The real expertise is that people who went through it, it’s embedded in their neurons. And it reminds me, you know, there’s this great statue in Rome by Bernini, and it’s an obelisk, which is the sort of Egyptian encapsulation of knowledge, right? It looks like the Washington Monument and the obelisk sits on the back of an elephant. And Benigni’s story was that knowledge should always be supported by wisdom. And I feel like when we see these models now, it looks like knowledge, but it’s lacking wisdom. And the reality is when, you know, there’s the obvious laughable hallucinations and then there’s the credible hallucinations. The only way to tell the difference between the credible hallucination and the thing that is actually good to go is for somebody to have the actual experience in their brain of like Noah, I know this. And to have the ability to say, ooh, that doesn’t look right. And I don’t know, but let me go ask Swap, because I know he’s an expert in XYZ area. And so I think we have this phase where it felt like expertise no longer matter, because I could just ask the bot. And I think we have to remember that that is only the illusion of expertise. That is only a distillation of things that have been written. It does not replace actual expertise. Because if you rely on the thing that’s synthetically generated as your only source of the expertise, and you know, there’s hallucinations and miscalibrations, you’re going to end up in a world of it. And so, you know, part of it is this whole reluctance to hire new people and bring new insights again, until we get to something like true artificial general intelligence, I just want, I constantly warning people, beware of the illusion of expertise in the machine. It looks great. It can help an actual expert be incredibly efficient. We heard Jim talking this morning about Greg using him. It is extremely dangerous to put it in the hands of a naive user and expect that things will not go wrong.

Swapnil Bhartiya: We’re now listening to another challenge is that when we do look at that so called expertise, what AI agents are referring to is actually created by humans. So we have that set of knowledge what after that, because AI agents will regurgitate, they are not going to create something. You know, it’s very, very challenging. So we have to continue to keep filling up the pool or the pond, not just keep draining it. So humans are, and also we are human civilizations, so we have to worry about humans in general.

Clyde Seepersad: To me the most accessible example is that containers were one of the early things in the Linux ecosystem back in the early 90s and we went for 15, almost 20 years before somebody had the insight that wait a minute, containerization is virtualization. Well, the agents would never have gotten to that because there was nothing in the entire corpus of published material saying could I use a container as part of virtualization strategy? Like a human had that insight and now we built the entire cloud infrastructure on top of it. Right. And so just because the agent has read everything ever written is not the same as insight and wisdom about what to do next.

Swapnil Bhartiya: Well, so true. And the more we realize that, the better it is for us to understand that the, the utopia that we are thinking and dreaming of it might never arrive. You know, it will. I always feel that and I, I think that should always be the case that humans will remain the operators and AI will remain the tool. It will never be the reverse. So, and I think that also paints a very good picture for where the job market is because there are panic. But I think after listening to you folks realize that things are not as bad as yes, things are changing, that is for sure. But there’s no need for panic. What is needed is for you to be prepared for you to understand where the world is heading because there may actually be a better position in the AI driven world because we will need more humans. AI will do a lot of low level things, but we will still need somebody to sit on top of that to keep an eye on it. So yes, there will be jobs. So Clyde, once again thank you so much for bringing all these great insights to kind of also contain some of the fear, panic and also being able to relate to where people are versus just talking about high level vision that hey, this is how things look like because that reality check is very important. You gave example of small and medium business, which is true as well. So and I also remember your example from last year about plumber that you know, so I was actually reading that a lot of billionaires. They are setting up to those vocational jobs to train people on those jobs, because those jobs will pay more. And those jobs are the one which will run the world, not the AI. Once again, thank you so much, and I look forward to chatting with you.

Clyde Seepersad: Always a pleasure. Thanks for having me back. Swab.

Swapnil Bhartiya: Awesome.

Clyde Seepersad: Thank you.