Rochana Golani, VP of Learning and Enablement at Databricks, on training people for roles nobody can name yet.
In episode 4 of The Learning Curve, Rochana Golani, VP of Learning and Enablement at Databricks, tells Vocareum CEO Sanjay Srivastava how you train people when nobody knows what the roles will be. She explains why AI fluency is now expected in every job, why assessing who sits at its cutting edge is a myth, and how Databricks infers employees’ skills from the work they are already doing.
Rochana Golani has built technical learning at VMware, AWS and Google Cloud. She has led learning and enablement at Databricks since 2022.
About the guest
VP of Learning & Enablement, Databricks
Golani has spent about 25 years building how people learn technical skills, at the companies defining each wave of infrastructure. She led training and certification at VMware, built a global curriculum and certification program at AWS, and ran learning services at Google Cloud, reaching millions of developers. She joined Databricks in 2022, two months before ChatGPT was released, and now runs learning for employees, partners, and thousands of enterprise customers.
Golani treats the expectation as settled: “it doesn’t matter which role you are in… absolutely every role needs to be able to be AI-fluent.” Measuring it is the part she rejects, because the test decays faster than you can administer it. Six months ago she would have been assessing whether someone could max out tokens — “which is the wrong thing to do.” What she measures instead is how much value someone drives for the smallest amount of money.
The cloud era ran on persona tracks — Solutions Architect, DevOps — because everyone could see the destination. That model is gone: “we don’t know what the roles of the future are yet.” Databricks rewrote its own enablement career ladders around three criteria instead: can they use AI, can they think critically, can they adapt fast. Her summary — “the ability to learn fast is probably your single most important skill.”
Databricks infers employees’ skills on the fly from the work they are already doing, then recommends learning from that activity — reading how a seller runs a discovery call, or how someone architects in their workspace. Assessment stops being a separate event in a separate environment. Golani says this is realistic for most companies, not just data companies: if you have the data, building inference on top of it is possible.
Rehearsing a pitch to an executive has always been limited by executives being scarce. Databricks rolled out AI practice across its whole field organization — not as good as the real thing, but available to everyone, with feedback attached. Srivastava’s counterpart finding from Vocareum’s schools work: AI did not make students better at math. Practice did. AI lowered the barrier to practicing.
Rochana Golani  · VP of Learning & Enablement, Databricks  · 10:16
Edited for clarity and length. Substance unchanged.
Sanjay SrivastavaWelcome to The Learning Curve. My guest today is Rochana Golani. Rochana has spent about 25 years building the way people learn technical skills, and she has done it at the companies that define each wave. She led training and certifications at VMware, then built a global curriculum and certification program at AWS, then ran learning services at Google Cloud, reaching millions of developers. Since 2022 she has been VP of Learning and Enablement at Databricks, where she runs learning for employees, partners, and thousands of enterprise customers. If anyone has watched up close how the world actually acquires new technical skills through the cloud era and now the AI era, it's Rochana. Rochana, welcome to The Learning Curve.
Rochana GolaniSanjay, thank you. Absolute pleasure to be here today. I feel really grateful for the opportunities I've had at companies like VMware and AWS, Google, and now at Databricks — each one really the next wave of disruptive technology. But with every disruptive technology comes the need to develop new skills. And so I've had the opportunity every time to do that in a fresh way.
Sanjay SrivastavaYou've watched not only the disruptive technology and the need for learning, but also how there have been massive disruptions in how technology skills are acquired — especially cloud and AI. I was talking to Michael Horn, my last guest, and the theme was how the hope for cloud and AI is still playing out. So how do you think the whole acquisition process has changed because of the technology we can now use to deliver and acquire those skills?
Rochana GolaniThere are three thoughts that come to my mind. At the end of the day, the first thing that really drives learning is the need itself. There is this idea that something has changed, I need to acquire the skills. The individual has an intrinsic motivation to say, I need to acquire new skills because I need to get a better job, or the job that I have right now is not going to be the most lucrative in the future.
Rochana GolaniIf I think back to when we were building the Cloud Solutions Architect certification — prior to that, people were system administrators. And we knew that the admin job is not going to be the most successful one. How do you think about cloud architecture becoming the new thing? And so people wanted to develop those skills.
Rochana GolaniWhat was also interesting is that when you're thinking about infrastructure architecture skills, like we were doing in the VMware days or in the cloud days, the production environment for those technologies is a very expensive thing. So you're not going to learn in a production environment. You needed to do that somewhere else, get certified, and then come do it. But you still needed hands-on expertise to be able to do that.
Rochana GolaniWhat's changing for us now is that we have access to more technology to do more enablement. With cloud — even if you think about your own company, Sanjay, Vocareum — it's a cloud-based solution that was deployed to do hands-on enablement. We didn't have that in my VMware days. And now what we have is the power of technology to do enablement itself, which has really uncovered so much more for us. To me, you don't do enablement for technology without leveraging the technology itself.
Sanjay SrivastavaAbsolutely. You're talking about the need to go learn, become a solutions architect instead of an admin, because that's where the need will be. And I'm assuming there's a virtuous cycle — not only do the learners want to learn it, but the companies support it. They have a vested interest also, right?
Rochana GolaniExactly, 100%. The employer perception of all of this is also a huge driver in why people do it. One of the largest sources of learners that we got when we were in cloud was with our system integrator partners. What they are trying to say is, we have to do something differently for our customers. Or even the large customers that truly went from on-prem to cloud needed to upskill their own organization. So we would get hundreds and thousands of people who are all learning from the same organization, because they need to work on different kinds of projects or they need to upgrade their skills to something new. So this idea of what is the future role that I need to do — at least in the cloud days, it was very particular, very much about persona-based learning. The solutions architect track or DevOps track became a very important thing, because everyone was like, this is the new role and I need to be headed there.
Sanjay SrivastavaThat's pretty fascinating. So the whole idea that you're learning for a role — that was a relatively new thing.
Rochana GolaniI do think it started a while back — system admin tracks or database admin tracks were there. But this idea of a future role, a role that I want to have, and then doing learning on that, is, unfortunately or fortunately, I don't know, not true anymore in the AI world. Because honestly, we don't know what the roles of the future are yet. So we have had to rethink entirely how we would structure learning, how we would go about doing the things that we need to with this AI transformation.
Sanjay SrivastavaLet's talk about AI. Our customers are companies like Databricks and AWS, but also a fair amount of universities. And right now there's a huge tension, from both the provider as well as employers like myself. As an employer, I have this tension that I want the employees who are walking in the door to have deep fundamental skills. At the same time, they should have deep AI skills, because — as Ethan Mollick calls it — the jagged edge of AI. I want people to have lived on the jagged edge, ideally in the learning setting, because when they come to work they should have the deep expertise and the experience to say what AI can do and what AI cannot do at the edges. That's where companies like us have to go out there and create value. But the whole thing is being disrupted, because there's a huge stress that the traditional architecture of learning, where you do assessment, is being challenged. Where do you think it goes, and how does it change the context of customer education and corporate learning? How do you make AI have the right constructive role in the learning process?
Rochana GolaniWow, that's a very loaded topic, Sanjay. So let's take this one bit at a time. The first thing for me to acknowledge is that employers expecting AI fluency as part of every job role is a fact. It is not that some companies are doing this and others are not. Every company, absolutely — it doesn't matter which role you are in. Maybe some are developing, maybe some are already there, but absolutely every role needs to be able to be AI-fluent.
Rochana GolaniThat said, assessing if somebody is at the cutting edge of AI fluency is a myth. Because if I assessed somebody six months ago, I would have been assessing them on: can they max out tokens? Which is the wrong thing to do. Really what I want everyone to understand deeply is, in the smallest amount of money, can they drive the maximum value with AI? So productivity measures or proficiency measures are changing on an everyday basis.
Rochana GolaniAs an employer, we are very deeply focused on this idea that you want to have people who have the expertise for whatever we are asking them to do — a seller in selling, a solutions architect in getting technical wins, a professional services consultant in creating the best architectures and implementing for their customer. But can they do it faster, better, and just differently when they use AI? And so we expect AI fluency as part of that job.
Rochana GolaniBut we are also looking for people's curiosity to say: in exactly a month, when this looks different, can they adapt fast enough to the new thing? And really, because we are in this transition time — I might really sound like a learning professional here, Sanjay — but I really think the ability to learn fast is probably your single most important skill.
Sanjay SrivastavaThat makes a lot of sense. And if you're providing learning, then how do you provide that evidence to the employer and not leave all the burden on the employer to assess that? I think that's the challenge we are going through. I don't know if you saw that University of Chicago Law School just last week announced their whole AI policy. It's pretty detailed about which classes you can have it in, which classes you can't — it sounds like something which will change in a month.
Rochana GolaniIt has to be written much more abstract. Databricks recently, even in my team, went through our own career ladders for people in our team, enablement professionals. And we said, we need them to be AI fluent. What does that mean? So we're not being prescriptive in how they use AI, but: are they able to use it? Are they able to think critically? And can they adapt and learn fast enough? That's what we are after.
Rochana GolaniAnd I think back to something that you started talking about. What AI is also giving us the power to do is the ability to really assess skills while people are doing. I don't think learning and assessment need to happen in an environment that's different than the job itself. So one of the fundamental things that data combined with AI gives us is the ability to assess skills while work is being done. For example, very basic: I can look at how a seller is talking to the customer and figure out, do they have good discovery skills? I can look at people's workspaces and say, do they know how to architect solutions better? So a lot has changed with what capabilities and technologies we have in how assessments can be done also.
Sanjay SrivastavaThat's really powerful. The whole idea about assessment being a separate activity from working or learning hopefully is going to change, because now it's so much cheaper to assess while we are working or learning.
Rochana GolaniThere was this whole notion in learning — we've been fans of this idea that you learn by doing. So if I'm teaching somebody how to be a cloud architect, I expect them to go into the AWS environment, spin up your VMs, configure the storage, configure things very clearly. And I want to be able to see that they can do it. And now I'm saying that does not need to be learn by doing. I'm saying learn while doing. So I can assess what is happening and make the learning experience much more nudge-based, in the flow of work, and hyper-personalized. Because the cost of creating content, which was the long pole in the tent for enablement, is much cheaper now and much faster.
Sanjay SrivastavaIt has been pretty fascinating to watch. In Vocareum of late, we have been deploying this really deep domain-specific tutor. So something can actually go to your AWS account and watch what you are doing and just suggest, yeah, you made a mistake here, let's go focus on that.
Rochana GolaniThat right there is the learn-while-doing idea. Actually, 2022 was when ChatGPT was introduced, two months after I took this role at Databricks, and literally the whole world changed. We are in the gen AI era, and we were all trying to think about what this could mean to learning. One of the big ideas that I think we stuck to then was that as learning professionals, we've always dreamed of this idea that we want learning to be personalized. Even today in university, student-to-faculty ratio is such an important thing for people. And I think it's important because of this idea that each student can have a bit more personalized attention from the faculty. If you bring that to the digital world, what AI and data now allows us to do — with the example of the AI tutor that you were giving — is make learning much more personalized for absolutely everybody. Because you can do that in the flow of work while they're doing it, and hyper-personalized.
Sanjay SrivastavaEvery time I have talked to you, you have always been a big advocate of personalization. That kind of a promise now, right?
Rochana GolaniYes. And we are actually doing this internally at Databricks, very happily so, I want to say — and very proud of the team that has pulled this off. We do the skills inference on the fly for where people are and what they're doing, and then make training recommendations or learning recommendations based on exactly the activity that they have done. So we are not doing assessment as a separate activity, but assessment of skills while the work is being done.
Sanjay SrivastavaThat's amazing. And obviously, Databricks being a data company, being able to collect all this data and process all this data — I'm sure you guys are ahead of most of the market.
Rochana GolaniYeah, but I genuinely think it's possible for most companies. If they have the data, then creating a skills inference on top of that and making that experience much more personal for users is really possible. Like I said, I think it was something that we all dreamed of as learning professionals, but it's actually possible. Hyper-personalized, and assessment does not have to be a separate activity.
Sanjay SrivastavaAnd the recommendation happens on the fly. When you say, okay, I'm going to go talk to my customer in two days — I need what I need right now. And now it's possible.
Rochana GolaniCorrect. Just extending the thought that you have: let's say I'm a salesperson and I have a meeting coming up with some chief data officer from company XYZ in two days, and they want to talk about product Y. I have never spoken about product Y before, because that's not in my skills engine. What if the tool that we are talking about can look at that and put an hour on my calendar the day before and say, why don't you practice a pitch for what you're going to do? And it spins it up, pretends to be the CDO of the other company, and gives you feedback. So you have actually practiced before you did your work. And it's very customized to the thing that you want it to do.
Sanjay SrivastavaThat's a great example. You're not only practicing the technical skill, but also how you communicate it. You're getting feedback at the same time, hyper-personalized for both you and the context.
Rochana GolaniThere's data that says people don't spend enough time in practice, which would have led to much better outcomes. And if they did that, then that could change the outcome of that one meeting that they probably wouldn't get a second chance for anyway.
Sanjay SrivastavaOne thing which I do not want to forget to ask you: I remember last time you were talking about the story of a teacher who personally made an impact on you. I just want my audience to hear about that story if you don't mind.
Rochana GolaniYeah, I hope she's also listening to it. The story I was telling you about was how, when I got out of college, my very first job was a trainer job. So there was nothing that I had earned in terms of experience or credibility in actually talking to enterprise customers about whatever the database and networking technologies of those days were. And I was privileged enough to have a mentor who happened to be my manager in this company. She was this amazing woman who physically was not very tall — I was much taller than her, but I always looked up to her — and she was so inspirational in just how she had control over the room, how she managed everything, and how she inspired me to say, I think you can do this. That confidence in someone who was fresh out of college, from somebody else who clearly knew what they were doing, I think helped me be on this path. And I'm incredibly grateful for that opportunity to meet with her, and then to be able to make a run for my own career with that inspiration from her.
Sanjay SrivastavaI have always been saying that your first encounter, your first manager, your first mentor can totally make or break your career. I was pretty lucky in my career too — I had just an awesome mentor. It's a great story.
Rochana GolaniSanjay, my question back to you then. You've also seen this from a lot of different angles, having your own experience with universities, but also companies that you founded and run so beautifully, supporting us as well at Databricks. What's your take on what AI can change for education?
Sanjay SrivastavaThere are a lot of vectors. Of the things which we are personally excited about, I would say three. One is what you mentioned already — in corporate learning, how do we play our part, where people just learn the job. There are two other things for us. One, for the math, for the students coming in California schools, we built a system which dropped the failure rates by two thirds. And what was really interesting learning there, now that we had all the data, and it's actually broadly applicable in learning, is that AI did not make people better at math. It was actually the practice which made them better. But it was the AI which lowered the bar for people to practice.
Sanjay SrivastavaAnd one of the things we are doing research on is: if you provide AI support — I'm talking about colleges now — the scores go up, but if you put them in an at-home or in-person exam, their scores will go down if we take the AI support away. So the system that our collaborators built, the researchers, and we were just implementing it: Sunday you had AI support, on Monday homework, Wednesday you did not, and then Friday you had to do a test. That alone, that progression, making sure that the AI support — again, this is for math, this is a different beast.
Sanjay SrivastavaAnd the last part is exciting. Some of our partners are saying, I will take AI to substantially increase the funnel. I can have just the AI teacher there. Maybe it's not as effective as also having a person, but by lowering the cost, now I can reach millions more learners. And if I get some of them into the funnel so that they go out there and have social mobility — those are some of the things which I'm seeing AI do, and there are already concrete results.
Rochana GolaniI think this lowering the cost of practice is such a powerful thought that I just want to double down on and give our own example. As learning professionals, we always knew that for communications or presentations, public speaking practice is really hard to do — because if I want to practice how to speak to an executive, the executive is a very hard resource to find. It's not like I'm going to find many executives to practice with. But the ability to be able to do that with AI just lowers the bar. And we rolled out this program within Databricks to actually do that for all of our sellers, for all of our field, actually. And I'm not saying that that was just as good as speaking to an executive. There was still an AI. But it definitely raised the bar for everybody in giving them the practice and giving them the tools and live feedback. That's another thing you never get — real feedback on what you can change. And so we were able to do all of that by leveraging AI in a very meaningful way for the organization.
Sanjay SrivastavaWe have created a tool called Atlas. Now you can create a simulation within 10 minutes — a nursing simulator, a virtual patient. I want to have a difficult conversation while I am giving them a recommendation. I can create that in 10 minutes. So you can practice, practice, practice so easily.
Rochana GolaniSo easily. Tough conversations for managers with their employees — those were things that became really difficult, and now somebody can give you real feedback on what else you could have done differently. So I think the opportunity as an enablement professional now, to be able to do all the things that you dreamt of, all the things that you read in a great research paper about personalized learning and in the flow of learning, I think are material now. You can actually do them, and they are not just good ideas — they can be implemented. The cost of implementing good ideas has come down.
Sanjay SrivastavaIt has come down a lot. And while the content creation was becoming easier, the value of the experiential learning is going up — but now you can create those experiential learning environments and feedback so much easier. And I think in higher ed there's always kind of friction of deployment. It takes a long time, because of all the good reasons. But in the enterprise world, things can happen much quicker, much faster. And so you are basically implementing the change in real time. That's pretty exciting.
Rochana GolaniAnd honestly, Sanjay, one other thing I will say is that when you're going fast and when you're early adopters, things go wrong as well. In this effort to go fast and try things new, we've had our fair share of problems. So it's not been as rosy all the time. We rolled out this massive program for a very large event that we did, and the model — or the tool that we had, the AI tutor — was hallucinating at one point in time, even though we had done all the testing, and broke at scale. So there are still those challenges. At the end of the day, technology is still finite and it still has a long way to come, but we are able to do way more things than we did before. And so we've had to learn how to pivot quickly as well.
Sanjay SrivastavaRight. And especially for the scaling, there is no other way to do it than just do it. I learned the lessons of scaling.
Rochana GolaniMultiple times in my career I have had that too, where you tried something and you scaled and then it just completely failed, and then everybody's running around like a chicken with their head cut off because we don't know what we're doing. But you need to be able to take those risks and go fast. And that's how you figure out the next best path.
Sanjay SrivastavaI think we are towards the end of the podcast. Rochana, any other final thoughts for our audience on where you see this going?
Rochana GolaniI hope there isn't any learning organization that isn't already experimenting with some sort of AI in their tool set. But most organizations start off with just saying, we're going to create more content because we can, or we're going to create content faster. My message to everybody there is that content is not the be-all end-all. Unless you change the experience, creating more content alone is actually going to help nobody. It just becomes — or it feels like — AI slop, even if it is really great content. So my message to all learning leaders is to start with the learning experience and let the content follow, versus start with content. Keep the focus on who you're trying to help, the learner at the center of everything that we are trying to do. And the learner experience, when that changes, I think the engagement from the user changes and content will get better. But starting with content has left us in a lot of difficult situations. So from my own experience, I would say focus on the learner experience first.
Sanjay SrivastavaThank you, Rochana. Great closing thoughts. Thank you once again.
Rochana GolaniThank you, Sanjay.
You stop treating it as a fixed proficiency test. Golani looks for value per dollar, critical thinking, and curiosity — whether someone adapts when the tooling looks different in a month — rather than mastery of a toolset that will have moved by the time the assessment is scored.
Deriving what someone can do from the work they are already doing rather than from a test — reading how a seller runs a discovery conversation, or how someone architects in their workspace. Databricks runs it on the fly and generates learning recommendations from the activity itself.
Srivastava’s answer from Vocareum’s own deployment is a useful correction: AI did not make students better at math. Practice did. The program staggered support deliberately — assistance on Sunday’s work, none on Wednesday’s, a test on Friday — so the gain survived the removal of the AI.
No — Golani’s closing advice is the opposite. Teams that lead with content volume produce something that reads as AI slop even when the content is good, because nothing about the experience changed. Start with the learner experience and let the content follow.
Conversations on AI, learning, and workforce readiness with leaders across education, technology, and industry.
