Untangling the “Cosmic Coincidence” of the Job Market
How do people find good jobs? How do employers find good talent? How should the American economy prepare for the future of work in the age of artificial intelligence?
On this episode, Issues editor Molly Galvin talks to Burning Glass Institute president Matt Sigelman about these questions. Sigelman spent two decades developing the field of real-time labor market data as CEO and then chairman of a company now known as Lightcast. Five years ago, he founded Burning Glass to transform the way that policymakers, researchers, employers, education institutions, and workers understand, plan for, and connect with the world of work.
An edited version of this interview appears in our Spring 2026 issue.
Resources
- Read the Issues Spring 2026 interview with Matt Sigelman.
- Learn more about Sigelman’s work by visiting the Burning Glass Institute and Lightcast websites.
Transcript
Molly Galvin: Welcome to The Ongoing Transformation, a podcast from Issues in Science and Technology. Issues is a quarterly journal published by the National Academy of Sciences and Arizona State University.
How do people find good jobs? How do employers find good talent? How should the American economy prepare for the future of work in the age of AI? I’m Molly Galvin, an editor at Issues. On this episode, I’m joined by Matt Sigelman to think about these questions. Matt is the president of the Burning Glass Institute, which is dedicated to using data to help us understand and plan for the future of the labor market.
Matt, welcome. Thanks for joining us. So what got you interested in the labor market and measuring employment?
Matt Sigelman: So as an undergraduate, as a student of public policy, I’ll spare you of the twist, turns, and tangles of early career, but simply to say that I wound up spending time thinking or wound up working in an array of problems around the connection between people and jobs and trying to understand how that can happen more effectively and efficiently. Initially, I wasn’t thinking about that as a market problem. I was thinking about it as more of the unit record problem of how do you match a person to a job?
So thinking about these as a defect in one-to-one matching rather than as market-wide problems. And so through work I was doing in that space—the private sector—developed a set of solutions to try to bring more efficiency to that one-to-one matching problem. But what I came to realize is that as long as you have a job market whose core motion is millions of individual random transactions, you’re going to have a market whose key motor is essentially cosmic coincidence. Think about all the times you go into, I don’t know, a restaurant and you see an employee who’s fantastic and you think, “What is she doing there? So many places she could be.” And what she’s doing there is that she happened to need a job and that restaurant happened to be hiring. And it was just this sort of cosmic coincidence that they got lucky enough to have that fantastic employee. And so I came to realize that the biggest tragedy, but also opportunity, of the labor market is that often people in opportunities are just a few skills apart. And so that becomes an information breakdown.
So I spent about 20 years building a company that’s now called Lightcast, whose key motor was to actually then say, whose key mission, rather, was to bring innovation to how we collect data about the labor market and where opportunity is, where talent is. And the company continues to do great work in that space. I left it about five years ago to start the Burning Glass Institute as a fully independent nonprofit data lab. Our work is shaped by the realization that is as important as great data sets are, at the end of the day, big transformations need to start with the problem.
Galvin: Right. Instead of the dinging.
Sigelman: Yeah. Where are the breakdowns happening in the intersection between work and learning, which is where we focus as an institute? What are the problems that, if you could solve them, would unlock economic mobility, would lead to better outcomes for workers and learners?
Galvin: How would you say that the methods that Burning Glass uses for gathering data compared to traditional indicators and how those data are gathered? How is that different?
Sigelman: So what I would say is this: is that traditional labor market data are survey-based and they’re survey-based for a reason. Some of that is historical. These were instruments that were first developed in the twentieth century when it would’ve been impossible to even conceive that you could actually capture data on the significant majority of labor market transactions. So as a result, you do relatively small sample-size surveys and you extrapolate.
What are the problems that, if you could solve them, would unlock economic mobility, would lead to better outcomes for workers and learners?
And so that’s why on the one hand, these data series are very valuable for being able to actually understand what’s going in the market. But they don’t lend themselves to effective practice or policymaking because they lack not only timeliness, but granularity. What we did instead is to say, “Okay, look, it is actually possible to see, to look at each individual job posting that’s out there. They’re all online, or most of them at least.” And we can see what are the specific things that employers are asking for. And to create essentially a language, a series of taxonomies for being able to define jobs with greater granularity for defining skills and being able to capture that in a more timely way. What we’re doing now at the Institute is to be able to say, “Okay, that’s all a really important starting point. How do we start to create an array of inferred or modeled values that go beyond just counting?”
It is helpful to be able to say how many jobs, if I’ve got a program to train people in Ruby on Rails, how many jobs are there and what other skills does someone need to know? But we also start need to be able to say, “Okay, well, when people get those jobs, what happens to them over time?” What kind of economic mobility do they achieve? Is that a good job? Is it a job that gives you good stability and good pay? Is it a job that gives you a foot on a ladder and good chances of moving up? Actually, the most people in that job wind up on a low wage treadmill. Probably that one, less likely in the last of those. But with a lot of the jobs, there hasn’t been that kind of transparency.
Galvin: How do you manage to track that, someone over their career?
Sigelman: So we’ve created a dataset that’s been particularly powerful in our work that’s comprised of the career histories of what’s now 65 million US workers.
Galvin: Wow.
Sigelman: Yeah. It is 40% of the US workforce. And so what it allows us to do is to say, essentially to think of it as a reverse RCT.
Galvin: Sorry, RCT?
Sigelman: Randomized control trial. So typically what you’re going to do in most rigorous studies, what you’re trying to do is sort of follow people over longitude. So okay, we’re going to see what happens and now I have to… I’ll come back in five years or I’ll come back in 10 years and I’m going to see what happened. For example, if you look at how colleges measure their outcomes, if they’re measuring post-graduation outcomes, it’s what they call first destination survey. Okay, where did where graduates land? Or the college scorecard, which is an excellent effort by the education department, tracks people for three years. But if I have your resume, if I have your LinkedIn profile, if I have any of a number of other things like that, what I’ve got is an ability to actually look back across the longitude of your career. And so I can look at things like, “Okay, at some point, Matt went to work at this company. How did that shape his onward trajectory?”
We’re doing a fascinating study right now: We’re looking at what happens to women who take a break in employment often for family care obligations. How does their onward trajectory compare with people who didn’t have that gap? And the wonderful thing about it is you can also start to create a broader set of metrics than has been historically possible, because you can look at not just how much people are earning or infer how much they’re earning based on the kind of roles they’re in. But what you’re able to do is look at other kinds of characteristics.
So if I want to understand… I’ll give you an example. We were doing some work recently with some HBCUs, some historically Black colleges and universities, and many HBCUs struggle with the same challenges that affect higher education institutions of all stripes. And so there are plenty of HBCUs whose post-completion outcomes are less than what you want. But it is also, something that we found in our analysis which was fascinating to us, that the share of HBCU graduates who wind up in a high social-impact field is much greater than the US higher-education student body at large. So it doesn’t explain all of the negative variants in wage, but certainly part of the reason why HBCU graduates in many cases are earning less is because they’ve made purposeful choices about to give back.
Galvin: Just to switch gears a little bit, I mean, you talked about AI before, and of course that is top most on a lot of employers’ minds and in a lot of employees’ minds. And there’s so many dire forecasts about how AI is going to affect jobs. Can you explain what you think the important dynamics are to pay attention to right now?
Sigelman: I don’t want to be dismissive of the potential for significant displacement. I think anyone who’s used AI, who’s used LLMs can see how ably they can take on some of the work that people are doing today. At the same time, I think the history of general-purpose technologies—whether we’re talking about electricity, whether we’re talking about automobiles, whether we’re talking about computers, whether we’re talking about the internet—actually doesn’t tend to suggest that those kinds of mass displacements are how things play out. What we see more often is that these technologies change the way we would work much more often than whether we work.
Take the internet. We recently did some analysis. We found that 90% of Americans work in an occupation which makes substantial use of the internet, whether you’re a web developer or a graphic designer, or whether you’re my barber who no longer needs to stop cutting my hair to answer the phone, but instead has an internet-based appointment system.
We’ve seen big changes in the nature of the skills that people are using, and we’ve seen very meaningful changes in the level of expertise that jobs require.
Now, relatively few people lost their jobs because of the internet. Mostly what we saw was the internet change the nature of the things that we could do. It changed workflows for how we do our jobs. And by the way, the job losses that occurred in many cases took a long time to happen. In fact, some of the ones that we’ve been seeing have really only been bearing out over the last several years. For example, it’s only over the last several years that we’ve seen a really big drop in demand for meter readers as connected devices have become much more of a thing over the last several years. You no longer need people to go walk around your house and have the dog bark at them while they check the meter. It’s no longer needed. Parking lot attendance, likewise, way down. There’s an app for that.
But this is not unique to the internet. It wasn’t the internet was a uniquely gentle technology. This bears out across the board, and it’s what we’re seeing in the data. Again, as I mentioned before, the data that we’ve seen thus far, there’s really very little evidence that despite the headlines of big layoffs that some companies are doing, but to date, we haven’t seen really big drops in employment.
We have seen changes in what happens inside jobs. We’ve seen big changes in the nature of the skills that people are using, and we’ve seen very meaningful changes in the level of expertise that jobs require, and it’s probably worth unpacking both of those. From a skill perspective, the sets of skills rather that would be reasonably predicted to be automated are 16% more likely to be in decline than skills overall.
Galvin: So you’re saying that that’s already happening, those skills are already declining?
Sigelman: That’s already happening. That’s an empirical observation of just looking at pre- and post-LLMs. Now, where it gets more fascinating is this: We have this kind of mental model of winners and losers, that some jobs that are going to be automated away, some jobs are going to gain AI superpowers. What we found instead is that the jobs that are experiencing the greatest levels of automation are the same jobs that are experiencing the greatest levels of augmentation. In fact, the correlation between those two is 0.87, if that means anything to your readers. It means that these are basically the same jobs. And so the actual mechanism is that as AI automates some tasks away, nature abhors a vacuum and the time gets filled with additional tasks.
Galvin: I see. So people’s jobs are really changing as these things become automated and taking on different tasks.
Sigelman: Those kinds of changes could be just as disruptive. Silent waters run deep, and the jobs may not go away, but if the skills that are required to do them change dramatically, it could still mean that how we prepare students to launch into their careers could be significantly disrupted. It could mean that workers who have been working in a field for decades, when they go to switch jobs or they lose their job, may find it very hard to compete for a new post because they no longer said to have the skills that you would need to be able to get a new job in the field. So those are really important effects for us to be able to track. We need to be able to see what is the nature of those skill changes role by role.
Part of the problem here is that we have a very industrial-era view of automation. In the industrial era, so much of what we sought to do was to simplify jobs—to make it easier to train people up, to be able to commoditize labor; essentially to de-skill them. And so jobs became stripped down to one or two tasks. If you automate the task, you automate the job. But jobs in the twenty-first-century knowledge economy are complex. Think about all the different things that you do with your time. So if you automate a task, you’re not necessarily automating the job.
Now, the other potential disruption though that we’re tracking is one around, is what we’ve come to call, my colleague at Harvard, Joe Fuller, worked together on an analysis of what we call the expertise upheaval. What we find is that jobs have different shaped learning curves. There’s some jobs where there’s some knowledge that has to be accrued upfront, and then after that, productivity doesn’t increase that much over time. Think about a bus driver. A bus driver, whether she’s been on the job for three weeks or for 30 years, her job is to show up on time, don’t crash the bus.
Now think about a data scientist and the difference in productivity between somebody who’s just getting started in her career and somebody who’s an expert data scientist. It is a huge difference in terms of their capability. In jobs like that, our learning curve is long and slow.
I’m going to imagine that most people reading this are people whose careers followed that long and slow model, that long and slow path. When you finished your education, your formal education, you were hired most likely more for your capacity than for specific capabilities. You were given humble tasks, you developed some proficiency, and then you were given somewhat more complex tasks. Eventually you developed some intuition and you developed expertise and you became good.
Now, in fields like that, when you look at, well, what are the tasks that you do at the entry level, what are the tasks you do is once you become an expert, the tasks that you did at the entry level are exactly the things that AI does really well. So it’s super tempting for employers to say, “Well, why would I want to hire somebody at the entry level when somebody who’s really good can really unleash the power of AI?”
And so it means on the one hand, it presents two really major sets of potential disruptions. One for employers: “Okay, where do I get experts if the way that people develop expertise historically is by starting a job?” And then it’s a big disruption for educators and for the students they teach because what it really says is that students are going to need to learn how to start their career in the middle instead of at the beginning. How do you learn to do a 22 what you used to not need to know until you were 27? And again, that’s not just a speculative finding. We see empirically in the data across the knowledge economy. Most occupations are seeing a significant shift away from hiring at the entry level toward hiring more expert talent.
Galvin: Okay. Well, that kind of leads me to another question I wanted to ask you. People are pretty pessimistic right now about getting a four-year degree and wondering if it’s still worth it. Are you going to get a good job when you graduate? Are you going to have a good income? What would you say about those concerns? I mean, you were basically talking about that earlier, but based on the analyses that you have done, is it still worth it to get a traditional four-year degree?
Sigelman: On the one hand, there has been no greater motor of economic mobility than the college degree, and that continues to be true. On the other hand, people are right to be skeptical. It’s not that the college degree isn’t worth, but it’s worth it too few people.
So think about this math. Of every hundred high school graduates, 66 go on to college. Of them, about 42 or 43 of the original 100 will graduate from college, and of them, only about 23 will land into a job they needed to go to college to get.
The obligation on educators today and on policymakers [is] to figure out how do we make sure that we are developing the student supports, that we’re creating the accountability for outcomes that ensure that college is worth it for more students.
So there’s plenty of data that’s out there that will say that the average college goer will earn an incremental $1 million, $1.5 [million] over the course of their career. It’s not wrong, but the average masks a huge amount of dispersion. There’s lots of people who are getting hugely more value. They’re earning much more than they would’ve if they had eschewed college and just gotten a high school diploma. And there’s plenty of people who are barely doing any better. 52% of college graduates a year out are working in a job they didn’t need to go to college to get, and that’s not a problem that sorts itself out.
So these are real problems. But I think the answer here is not, “Okay, college isn’t worth it.” It’s “what do we need to do to make college worth it more consistently?” And I think that’s the obligation on educators today and on policymakers to figure out how do we make sure that we are developing the student supports, that we’re creating the accountability for outcomes that ensure that college is worth it for more students.
Galvin: So you’ve studied regional economic development. We’ve long seen certain “winning regions,” especially in tech. What do they get right?
Sigelman: So I wouldn’t want to single out any specific region, but what I can say is this, that winning regions recognize that an economic development strategy needs a talent strategy and a talent strategy needs an education strategy. That may sound obvious, but it’s not. Most places have an economic development plan, but the notion that if there’s sectors that you’re really prioritizing, if there are industries that you need to win at, then you need to make sure you have the talent to drive those industries. That’s something that’s not an obvious realization.
And by the way, this means also recognizing that this isn’t just a warm body problem–how do we make sure we have enough workers in those roles? But those workers need to have the skills, not just of where those roles in those sectors have been, but where they’re going. And that means having an infrastructure for skill development.
My friend Mitchell Stevens at Stanford liked to talk about the imperative to move from being what he calls a “schooled” society to a “learning” society. We’ve got a society, our society right now has had a great achievement in having universal basic education, but the underlying premise is that that education you receive in your first two decades will serve you throughout a lengthening lifespan, that it will serve you in a labor market where 37%, according to our analysis, of the average US jobs skills turnover in five years.
So that’s not a good assumption. And so what it says is that if your region, in order for your talent base continue to be valuable, you need to make sure that you have an infrastructure for people to be able to acquire skills on the fly. You need to make sure that you have the employer engagement to make sure that employers are training people for the future of their sectors, and are thinking long-term about human capital, as being as much an asset as any kind of physical plant and machinery.
Galvin: Right. The Burning Glass Institute is devoted to boosting economic mobility. What is your level of optimism that the US can do this?
Sigelman: I continue to believe that we can see strong levels of economic mobility in the future and perhaps improved levels of economic mobility from what we’re seeing today. This comes back to the question of how do we help people build value in their own labor over time? How do we help people become more productive? And how do we recognize that the key lever for increasing the productivity of our workforce is learning?
Each job transition that people make is an opportunity for people to either move up or get stuck. And the way that you move up is by building new skills.
Each job transition that people make is an opportunity for people to either move up or get stuck. And the way that you move up is by building new skills. You’re building on the skills that you already have, you’re filling gaps, you’re taking jobs that require new skills that ask you to be more productive. How are we helping people build those skills? How are we helping people grow their productivity? And how do we reshape our own understanding of productivity?
Right now, we tend to think about productivity as a way of reducing cost. Productivity is a ratio. It’s the ratio of the value of labor’s output to the cost of its input. And the reason why many of us flinch when we think about productivity, about productivity enhancement, is that productivity is almost always discussed in the context of the denominator of that ratio, of how we make people’s work cost less. How do we use less labor? How do we use lower cost labor? How do we automate labor away? We almost never are thinking about the numerator, which is to say, how do we make people’s work worth more?
And what will decide whether America sees a revitalization of the American dream and increased economic mobility is whether or not we see this moment, this AI moment, as being a time whose imperative is focused on making people’s work worth more, about helping people acquire the skills that improve their productivity and create greater value, not only their employers, but for themselves.
Galvin: Check our show notes to find links to the interview and more of Matt Sigelman’s work. Please subscribe to The Ongoing Transformation wherever you get your podcasts, and write to us at podcast@issues.org. Thanks to our editor, Shannon Lynch, and producer, Kimberly Quach. I’m Molly Galvin, an editor at Issues. Thank you for listening.