In Conversation with Josh Bersin: Why 2026 Changes Everything About How We Work

Introduction

Nolan: Hello everyone and welcome to the Talent Equation podcast. This episode is sponsored by Infopro Learning. As always, I’m your host, Nolan Hout. Joining me today is Josh Bersin, who probably does not need an introduction, but I’m going to give you one anyway. If you’ve worked in HR, talent, or training space at all, you’ve probably quoted Josh Bersin in a meeting. He started Bersin and Associates back in 2011, sold that to Deloitte, and has spent the last two decades shaping how the world thinks about work, learning, and talent.

His team just launched Galileo Mars, which is a big leap from the original Galileo product. In short, they’ve transformed what was already a powerful AI agent into something called a super agent. In today’s episode, we’ll learn more about Galileo Mars while also getting Josh’s view on where the future of our industry is headed. Without further ado, welcome to the Talent Equation, Josh.

Josh: Thank you, Nolan. Great to be here. Lots to talk about.

Josh Bersin’s Origin Story

Nolan: Before we get into Galileo Mars, for those who don’t know your story, can you take a moment and explain how you got into this field and how it’s led you to where you are today?

Josh: I inadvertently tripped into this wonderful profession in 1998 when I was leaving a tech company to work for a startup building an online learning system. At the time, the internet had barely been coined, but people were starting to do training over the internet. I spent a couple of years at that startup. The company was sold to a bigger company. I got to know a bunch of chief learning officers and realized there was going to be this huge transformation from instructor-led traditional training to online, which seems like it was so long ago now. The part of it I was particularly interested in was studying it, understanding the best practices, and writing about it.

So, I started getting involved in research, and that evolved into a company that, as you said, we sold to Deloitte. But around eight or nine years in, we realized the HR stuff and the learning stuff were interrelated. All the research we were doing in L&D was connected to much more going on in leadership development, succession, and everything else happening in the HR world. So, for the subsequent years, maybe ten to fifteen, we leveraged that research to study all the different domains of HR. And now it’s all about AI. So, we’re going through the reinvention again.

Nolan: What do you think made you stick with this? Curiosity, the lifelong interest in learning. At some point you could have taken a job as a CLO somewhere. What kept you in the research and writing side?

Josh: Well, first, there were no jobs in 2000. I was at a dotcom in the Bay Area, 9/11 had just happened, there were no jobs. But beyond that, this turned out to be a career I’m perfectly suited for that I didn’t know I could do. My father was a scientist. I grew up in a world where we were always asking questions like, why do you think the sky is blue? We were always thinking about understanding and learning.

My nature is to never assume the answer we currently believe is correct. There’s probably another way to think about it. In the human capital, training, and HR domain, there are no perfect answers to any question. Everything’s open to discussion. That’s kind of why I like it. And things repeat themselves in different forms over time. So, I’ve become more seasoned in understanding all these issues.

Why Josh Moved into AI Early

Nolan: When AI came about, you were one of the thought leaders in this space who jumped in early and recognized the potential. When did you launch the original Galileo?

Josh: About two and a half years ago. And let me explain why, because I think it will help people understand what’s going on. I spent years doing mainframe and PC work at IBM, then years at a database company where one of my jobs was running business analytics and data warehousing. I was learning about different forms of data, how data is manipulated, and multi-dimensional analysis. But up until AI, all the data in companies was siloed into functional areas or databases. Databases are structured, row-by-column storage containers, and they’re transactional. The problem companies have had since I was at Sybase, really since IBM, is: we’ve got all this data, who’s going to put it all together and make sense of it?

Then along comes AI and says, give me the whole data set and I’m going to vectorize it. Every piece of data is connected and indexed to every other piece of data. It’s a multi-dimensional system where anything that happens over here, we know the impact over here. Think about that in a business context where we have data about skills, jobs, safety, experience, college degree, all of it. AI sees all of it as one thing and understands the relationships between everything. And it can update itself through learning as the data changes. Suddenly, all that stuff I worked on for 25 years is a hundred times easier to get value from.

Apply that to L&D, where the basic model is: we’ve got a problem, we diagnose the learning needs, we teach people about the problem, and hopefully they get better. That’s a very old paradigm. But imagine the AI sees all of that and knows what good looks like because it can see the people and the content that perform best and can apply that to an individual. It’s so big and so fascinating that I couldn’t stop.

What happened for us was we had been publishing research in PDFs and videos for many years, essentially running a publishing company with an advisory culture. We were always frustrated because someone would call and say, Do you have any research on this? And we’d say we have fifty reports on that, but we hate to make you read them all. Well, now we don’t have to do that. When we put all our content into AI, it suddenly became intelligent about everything we’d done for thirty-five years.

It’s almost as if all the research ever done by all the analysts, some of whom don’t work for me anymore, is there for you to use. You can ask a question, or you can use the learning engine to generate courses, videos, podcasts, and checklists. I kind of saw this coming when I first heard about ChatGPT. Within three or four months, I saw the opportunity.

From AI Assistant to AI Agent to Super Agent

Nolan: When Galileo was originally launched, every company was facing the same decision: do you cannibalize your own business model or not? The ones who realized it was going to happen anyway are the ones who iterated the fastest.

Looking at this latest release, Mars, which came out around mid-to-late March, tell us what changed. There was a line I loved about moving from assistant to agent to super agent. What was that big shift?

Josh: When we first built Galileo, it was a really powerful assistant. You could ask it any question and it would teach you things or answer questions. But it wasn’t able to solve a problem. It could explain how you would solve a problem. What companies were doing with it was connecting it to their corporate systems and loading it with information like their skills model, employee lists, job titles, job levels, and salaries. The system was starting to prove it was a system of action, not just a system for answering questions.

So, in the Mars release, there’s a big new feature called workflows, where you can prompt and develop applications inside Galileo to solve problems. One of the workflows is called something like talent redeployment. It asks you a series of questions about what state you’re trying to move from and to, whether you’re trying to upskill a group of salespeople or move people from sales to consulting. Then, if you give it a list of job titles and details about the people, it will show you who should be trained for what role and what the development plans for each person are.

Clients didn’t believe me this was possible. It sounded too good to be true. So, we took data from their companies about their people, put it in, and it developed new development plans, showed who was most suitable for what role, identified skills gaps based on experience, and then, even more interestingly, we use Galileo to record all our company meetings too. When you throw that data in, it knows about people’s real skills and real work activities and makes even smarter decisions. We have about 400 of these workflows built out, and you can build your own.

Nolan: What you’re describing with just that one workflow is essentially what a fifty-million-dollar company does. That’s the beauty of a connected ecosystem. AI is consuming all these point solutions because it can release a new workflow that does what an entire startup does.

Josh: One of the things we did with Mars is we announced a version called Galileo for Consultants, because we also put in our consulting models and change management frameworks that we use with clients. It can now take a problem, as long as you can explain it and put the relevant data in, and it will diagnose the problem and step you through the solution.

And now that companies can see what this product can do, they want to connect it to their corporate systems. We have a version running on Microsoft Copilot, a version that works with ServiceNow, one that integrates with Workday, and a version for SAP. Depending on what kind of company you work in, you can get your company data into Galileo very easily and use the intelligence there to do these kinds of activities.

Data, Rule Books, and Making AI Agents Think Like Your Company

Nolan: Let’s talk about data for a second. As AI has matured, I started realizing the tip of the iceberg of the problem. At first, it was, if I give it a bad prompt, I get a bad answer. Bad data in, bad data out. Now that we have super agents like Galileo Mars, how important is the data, and how connected does it need to be?

Josh: The data is important because the system won’t make the right decisions unless it has the right data about your company and your people. The more data you give it, the smarter it will be. But the challenge isn’t usually the data itself, because most companies can get the data if they decide what they want. These systems are very good at understanding the structure of data without you having to tell it too much. If the data is in Excel or a flat file, it’ll make sense of it if you label it reasonably well.

What I think is more interesting is that the prompting or programming process is now much more like consulting. You give Galileo a task or a project, but you also have to give it a rubric of rules, because otherwise it’s going to solve the problem in whatever way it thinks it should. One of your rules might be: we can’t raise salaries by more than ten percent. Or: we can’t move people more than two hundred miles from their home location. Or: here’s our corporate leadership model with seven things we teach all managers, make sure you’re considering these in the solution.

And if you give it these rule books, the AI sort of becomes like your company. Where this is going over the next couple of years is the AI will start to tell you what the rubrics should be. It’s going to say: I’ve been observing what’s going on in your sales organization and you’re very quick to judge some of your salespeople when they need more time.

Nolan: Just yesterday, I was looking at a product I’m involved with called OneGuru, which is built around nudges. A lot of workplace technology still operates on a “tell me what I need to do” model. But as you said, once AI is embedded within a company and understands how the organization operates, it reaches a point where it can say, “You’re not hitting your targets this quarter. Have you considered taking this negotiation course?” At that stage, it starts to become a proactive agent.

Josh: One of the big L&D use cases that’s going to become much more common is this: we’ve got a bunch of people doing similar jobs, sales or customer service, and here’s a group of people who are killing their numbers. Super successful, great customer feedback, high revenue. What are they doing that we don’t know about? You can talk to them, go on a sales call with them, listen to them, but they don’t always know what they’re doing. They’re just good at it.

The AI knows. It can figure out the behaviors, the activities, how they’re spending their time. Things like that are going to be transformational. And then there could be an alert to the manager: you’ve got a bunch of people falling behind in these areas because we can see what the high performers are doing. So, it has the potential to become much more proactive.

Digital Twins and Ambient AI

Josh: In addition to all the Galileo stuff, we have another tool we call a digital twin. It looks at all the data in our Microsoft Outlook system and personifies each person’s emails into a persona. I can go online and have a text message conversation with Barbara, one of our salespeople, and ask: when was the last time you talked to a specific client? What did you talk about? Are there any open issues? And it responds as if it’s Barbara, based on her activities. She can turn off things she doesn’t want people to see.

I think this idea that AI is almost ambient in the company, that’s maybe a year or two away. But if you look at what Microsoft is doing with WorkIQ, which connects data and creates context, it’s going to be easier and easier for L&D tools to diagnose problems and implement solutions almost on their own.

Nolan: You talk about a big problem with skills validation. How do we really validate somebody’s skills? A big use case for that type of technology is interpreting skills by observing what people do. Seeing what they use in Excel, how they work, how they respond to emails, how they show up in meetings. The AI knows you, probably better than you know yourself, because it can identify your blind spots.

Josh: I talked to a CHRO the other day who records all his meetings and stores the data in his own local AI. Every Monday he asks his AI: what did I do well last week, what did I do poorly, and what could I do better based on what I’m trying to accomplish? He said the AI is like his personal coach.

Dynamic Enablement and the Future of the L&D Function

Nolan: You mentioned building courses faster as a way people think about AI, but you pushed back on that framing. What is the right way to think about what AI super agents actually change for L&D?

Josh: The phrase we use to describe where this is going is dynamic enablement. Many things we do in HR and L&D are really there to enable somebody to do something better, safer, more quickly, or more productively. We do that in an episodic, periodic way. We diagnose a problem, come up with a solution, and train people. Imagine if that happened in near real time by the learning and enablement platform, there to help you on a regular basis. Sometimes you ask a question and it answers you. Sometimes you say, I’d like to learn more, I have five minutes, can you give me a five-minute overview?

Or: I’d like to take an hour next week, book time on my calendar and teach me about this, develop me a thirty or forty-five minute course. That’s the way people actually learn. Learning is personalized. It’s different for everybody. Some people learn by talking to others. Some want a group discussion. Some want to sit and read. Some want to listen.

Google’s NotebookLM podcast is unbelievably good for learning. Something about two voices discussing a topic is incredibly effective. If you apply that to every enablement issue going on across a bank or insurance company or manufacturer, you’ve got this enablement challenge happening all over the company all the time. AI takes you from a traditional training paradigm where you’re building a solution and trying to personalize it, which is very hard, to a completely different situation where everything is personalized for your needs.

Nolan: When I hear people say AI will help me build courses faster, I understand the impulse, but that’s the old way of thinking about it. The course builds itself in real time by the user.

Josh: You can build courses as much as you want if that’s what your company needs. Sometimes people just want a course. But think of the word enablement more broadly. Am I in the right job? Is my manager the right manager for me? Am I working too hard? Is my schedule correct? Many of the things we try to optimize around productivity at work, that aren’t just training, have to do with enabling somebody to succeed.

And then what happens to the L&D team is a whole bunch of people who were locked in a closet building courses are now out there helping people enable themselves, watching what questions they’re asking, and saying: there’s a whole bunch of people in the company who don’t understand this topic, we need to build more curriculum, bring in more expertise, run a webinar. L&D will become more distributed, more decentralized, more embedded in the business. Many of these enablement problems are going to be handled by enablement groups within the business, not by a central L&D function. Either L&D gets redistributed into functional areas, or there will be enablement managers who feed the AI to make sure it’s current on the latest things happening in each domain.

What Talent Leaders Are Focused on in 2026

Nolan: As we look forward to wind up the podcast, I’d love to take a broader view. What are the big trends on the minds of talent leaders right now, the problems they’re trying to solve?

Josh: Right now, every company is wrestling with the same challenge: I’ve got legacy systems, Degreed, LinkedIn Learning, Cornerstone, Saba. What do we do with this relative to the new stuff? Do we wait for them to upgrade? Do we start over? I’ll give you our own example. We had an online academy that took five to seven years to build with hundreds of items in it.

When we got Galileo up and running, we wondered what to do with the academy. My original reaction was it would take at least a year, maybe two, to migrate everything. We did it in four months. I think there’s a lot of fear about moving to the new world, and some companies are moving much more slowly than they should. But as one client said to me recently: this is not a situation where you run a pilot and see how well it works.

This is more like rip off the bandaid, describe the future, and start going there now. In 2026, everyone is caught in confusion because vendors are throwing out near-perfect but unfinished products, and it’s not clear whether to go from an old platform to a native new one or keep what you have and wait for it to evolve. That’s very company-dependent on how much risk the company wants to take. But I’ll tell you: we pulled off the bandaid and we’re in a new world. We can iterate and grow so much faster now.

Nolan: The arc I’ve seen our company take mirrors what I see everywhere. There are the micro adoption and the macro adoption. It goes: this is cool, this sucks because everything it gives me is wrong, wait actually that was me giving it bad inputs, and now this is the greatest thing I’ve ever seen. I’m seeing that arc even within my own team. We just pushed everyone to a single enterprise LLM and it feels like the tipping point. You have to live in it. There isn’t really an option to stay on the sidelines. And the ones who get the most value are the ones who centralize their data and their AI in one location, because it will keep learning and getting better.

Josh: The other issue is the company itself. We were in Asia with a large US bank, showing people the new Galileo capabilities, and they were blown away. They said: we could do this, we could do that. I asked if they wanted a proposal. They said they couldn’t do anything until IT approved it. They couldn’t even issue an RFP yet. Other companies say: you come up with a good idea, and if it works, we’ll spread it around. There’s more focus on central decision-making now because the decisions are piling up. It could slow down a bit while some standards get established, but it’s happening very fast.

Closing Thoughts

Nolan: Josh, if people want to learn more about the Galileo Mars release and your overall research, what’s the best place to go?

Josh: There’s a whole website called getgalileo.ai with all the Galileo information, including a success center with videos that show how different things work. The main website is joshbersin.com. And we have a very highly subscribed podcast where I cover a lot of this, with tons of education about the market, AI, and L&D. Those are all good resources.

Nolan: Wonderful. Josh, thanks so much for joining us today. It’s been a pleasure.

Josh: Thank you, Nolan.