What we talked about
Madhuri Somara, Senior Product Manager at Microsoft, joins Federico to unpack how she builds AI agents that actually help people, not just impress on paper. Fresh off being honored with the 2025 Product Leader Award by Products That Count, Madhuri traces her path from coding and business analysis to product leadership, and why empathy, rigorous evaluations, and clear user value are her north stars.
Show notes
Madhuri Somara did not land at Microsoft by waiting for the right job posting, she spent years studying what Microsoft PMs actually did each day, building relationships on LinkedIn, and deliberately positioning herself for the role. That same intentionality defines how she builds products: before shipping the case management agent she owns, she shadowed support engineers to watch where they got frustrated, then tracked the sentiment shift when the agent removed that friction. The 2025 Product Leader Award from Products That Count followed.
What we covered
- Madhuri started her career as an AngularJS developer doing double duty as a business analyst for a government client in New York, which gave her early exposure to both sides of the build-and-ship process. She realized she was enjoying the customer conversations more than the code, and that realization pulled her toward product management.
- At GS1 US, the organization behind global barcode and data standards, she built a Power Automate flow that auto-filled a few fields in the customer service team’s workflow. What felt minor to her generated thank-you messages from the support engineers. Shadowing them afterward showed her that removing even two clicks from a repetitive workflow is a meaningful intervention in someone’s daily life.
- The case management agent she owns at Microsoft automates the full lifecycle of a support case, from creation through troubleshooting, resolution, and follow-up, for large enterprise customers. She emphasizes that humans can oversee and intervene at any point; the agent acts independently without requiring human approval at each step, but supervision is always available.
- AI evals, she argues, are now as non-negotiable for shipping AI products as testing is for software. Because agents learn and adapt rather than following fixed instructions, intuition-based testing is no longer sufficient, structured evaluations and golden datasets are what give teams and customers confidence that the model will behave as intended.
- On the tension between AI capability and responsible use: Madhuri’s guiding principle is to first ask whether a problem can be solved with or without AI, then decide whether to add it. She explicitly pushes back on the “everything is a nail” mindset she sees spreading, where AI gets forced into features that do not need it and ends up overcomplicating rather than improving the product.
- Her advice for women and underrepresented voices entering AI product leadership: network actively on LinkedIn, there are more mentors willing to coach than most people realize, and be genuinely curious, asking engineers how they built a model or asking designers what drove a particular decision, because that curiosity compounds over time in ways that cannot be taught.
About Madhuri
Madhuri Somara is a Senior Product Manager at Microsoft, where she leads AI agents, NLP, and intelligent automation initiatives including the case management agent announced at Microsoft Ignite 2024. She serves on the editorial board of Women in AI Ethics and was named one of the outstanding product leaders of 2025 by Products That Count.
- LinkedIn: http://www.linkedin.com/in/madhuri-somara-589375b2
- Website: https://microsoft.com
Episode 74 of the PreVetted Podcast.
Full transcript
Federico Ramallo (00:00) Welcome back to the PreVetted Podcast, where we spotlight extraordinary people and remarkable talent reshaping our world. Today we are joined by Madhuri Somara, she’s Senior Product Manager at Microsoft, where she leads AI agents, NLP and Intelligent Automation Initiatives. She has over a decade of experience. She has built a career at the intersection of advanced AI technologies,
and user-friendly applications. She has played a key role in launching transformative generative AI solutions, including the high impact autonomous agents announced at Microsoft Ignite 2024. Most recently, Madhuri was honored with the 2025 Product Leader Award by Products That Count, recognized her as one of the outstanding product leaders shaping innovation and impact across industries.
Beyond her work at Microsoft, also serves on the editorial board of Women in AI Ethics, advocating for responsible innovation and diverse voices in technology. Madhuri, welcome to the show.
Madhuri Somara (01:10) Thank you so much, Federico. I’m excited to be here. And thank you for reaching out to me. I have been seeing your podcasts and I’m really excited and looking forward to this session.
Federico Ramallo (01:21) Amazing. I’m honored to have you here today. First, before we start with the questions, congratulations on the Product Leader Award. That’s a great achievement.
Madhuri Somara (01:30) Thank you, absolutely. And that’s a great intro. Thank you. Appreciate it.
Federico Ramallo (01:34) Awesome. So what first drew you to product management and AI? How did your journey evolve to Microsoft?
Madhuri Somara (01:45) Absolutely. That’s a great first question.
Before even product management, becoming a product manager, I actually started my career as an AngularJS developer for a government client in New York. I was an intern. So I was wearing multiple hats. Like I was a business analyst as well. So which kind of gave me exposure to both the sites. So I was writing the code and I was also talking to the customers actually. So during this process, I quickly realized, you know, that I was enjoying conversations with
customers more than actually writing the code.
kind of realization pulled me into the product management where I actually could blend that customer empathy, design ideas, the vision, and know, even looking at the data, you know, finalizing those product features or like, you know, prioritizing those product features. And then I, after internship, when I joined UpNod, I was much closer to this Microsoft ecosystem, like, you know, designing and delivering solutions on
Microsoft Dynamics and Azure, which I was working on even before that. So over these years, I developed an understanding how these product features are strategized, prioritized, and launched. So there’s a lot of effort that goes into it, right? So even with the vision and launching it. So after nearly seven years of working on that product strategy and product features, launching and everything of that sort,
I kind of, you know, it felt natural to me that the next logical step or next best step for me to is to join the mothership itself.
And I genuinely manifested that move. I built relationships in LinkedIn, reached out to Microsoft PMs, talked about what they do, understood what they do in a day, how they prioritize these features and everything like that. So I really positioned myself for the opportunity. was intentional in trying to be on the mothership. So yeah, today at Microsoft, I’m working on the AI agents for customer service. It’s like a full circle journey for me, absolutely.
Absolutely.
Federico Ramallo (03:48) Amazing, amazing. I think that when you have an objective, you can focus your energy until you achieve it.
I don’t know, I don’t recall if I told you the story, but I joined Microsoft when I was 16 years old and I had a similar situation that I was coached by people at Microsoft and I was like, well, that’s my dream. I want to join the company. Right. And, and then eight months later I was in the company, you know, and it happened so fast.
Madhuri Somara (04:21) I know.
Federico Ramallo (04:21) Yeah, so I can relate to that, yeah. ⁓
Madhuri Somara (04:24) Absolutely.
So you manifested joining the Microsoft as well. yeah, manifestation is powerful. Yeah.
Federico Ramallo (04:30) Yes.
Yeah, I mean, it was great. I had a great experience there. I learned a lot from a lot of people. This was like 20 plus years ago when Microsoft was going from being this crazy cool company to being more of a, know, serious company, quote unquote, right? Yeah, so they were.
Madhuri Somara (04:48) Yup, slowly.
Federico Ramallo (04:52) building more of enterprise policies and things like that. But I heard lot of fun stories, so that was fun. So looking back, what pivotal moments or projects shaped your career path the most?
Madhuri Somara (05:00) Yeah.
Good questions. Let me start with seeing this. I wouldn’t say.
there’s one particular moment that happened overnight. So my direction or my vision or my desire to be in product management became clearer with every project or even every person I met, every team I was on, every challenge that I faced. So during the internship process or in companies where I was working as a product manager.
this process I went from writing code to thinking about you know the customer you know how they want the product to work for them right. So looking back there’s not one moment that really shaped me so there there is a collection of moments. The first one that I could think of is for me you know deciding that I have to do my masters in US like thousands of international students you know chasing new possibilities. So it wasn’t just about academics for me it was about adjusting to
new culture, new work ethic or you know even selling my idea. It’s totally different. I learned a new way of selling my idea, getting everybody on board you know in the teams. So I learned all of that here and that helped to strategize my path and gave me the confidence to navigate the change like ambiguity around me. And there’s another defining moment where I was working with a mentor who helped shift
my mindset like you know while I was working with him I am working with him now he helped me focus on how to build a product from the customer’s point of view you know how to solve real problems not just you know making some fancy stuff and stuff like that so that perspective really changed a lot for me that’s how I approach prioritization and even decision making and even how I build the features today so yeah those are some moments that I could think of
Federico Ramallo (06:57) Yes, mean, sometimes we don’t know the good times, right, until they’re already past, right? Or we don’t have like one specific aha moment, but it’s the accumulation of all the decisions that we make that help us get where we are, right? So it’s more of an evolution than a revolution, right?
Madhuri Somara (07:16) Absolutely, you’re spot on. Now that you’re saying that, that’s actually true because when I was working on internships or when I was working in those companies, it wasn’t clear. I wasn’t realizing the value of learnings I’m getting from the companies and from my mentors. But now when I look back, it’s huge. That’s what shaped my career path. So yeah, you’re spot on.
Federico Ramallo (07:38) Yeah, this reminds me, this is a little tough topic, but this reminds me of Schwarz and Eger in an interview he was talking about on Terminator. He wanted to change the phrase from I’ll be back to I will be back, right? Because he thinking is like, well, machines don’t use contraptions, right? And eventually, you know, there was a back and forth and it was decided to I’ll be back. But at that time,
he didn’t think much of it other than that. But then people pick up that phrase and that kind of resonate with people, right? So he was saying, sometimes you say phrases that you don’t know which one are going to resonate with people, right? So, and that kind of stick with his career, you know, because people ask him.
to say I’ll be back all the time, right? So anyway, I think same thing happens, right? I mean, you work on projects and you don’t know how that’s going to affect the future of your career, right?
Madhuri Somara (08:29) Really? Yep.
Federico Ramallo (08:30) So can you share us a little more about your time at Avanade and GS1 US? How that prepare you for leading AI initiatives at Microsoft?
Madhuri Somara (08:40) Absolutely. those two are out of my career. Those two are my favorite companies as well. But both my time at GS1 US and Abnaud.
Federico Ramallo (08:46) you
Madhuri Somara (08:50) they played a huge role in shaping how I think about products now, products and people, you know, how to talk to customers or like how they are thinking to understand what customers want, right? Even for example, at GS1 US, which actually supports millions of businesses globally through, they’re the barcode manufacturers and data standards. So my job, my only job at GS1 US was to make life simple for the customer service team. And I once, I remember I once built
a power automate flow where it was just auto-filling a few fields. For me, it really seemed minor, but I started getting thank you messages from the support engineers from the customer service team. I was thinking, I just sat and thought, is this really such a big deal? It was pretty quick for me. Then I went on and shadowed them how they do things in talking to the customers and stuff like that.
then I realized it is pretty interesting or it is pretty impactful for them because that automating couple of clicks, removing the friction from their workflow means a lot to them. So even it comes from removing small daily pain points that makes people’s jobs easier, that’s a big deal for them. So that’s when I understood, even if you’re saving couple of clicks, it’s huge in that
particular workflow in the support industry’s daily life. And even at Uvnot, right? After GS1 US, I joined Uvnot and I worked closely with financial customers, Where data, accuracy, compliance, all of these are really important for them. It mattered so much for these, even now, we all know, right? For financial clients, it matters so much. Seeing how even a slide process
gap can cause a disturbance across the entire systems. That gave me an understanding or sense to take that precision or even how to prioritize those products or product features, to build around this data compliance standards and all of this. That gave me a deeper understanding of the systems. So all these experiences actually helped me gain empathy towards a customer.
I was on both the sides, customers and you know, who are experiencing the product and who are building the product, right? So, you know, this really helped me and that’s how I lead, know, or I work on AI initiatives at Microsoft today. It’s really helpful for me, yeah.
Federico Ramallo (11:17) Right, right, that’s interesting. I think that the key for product management is empathy, as you said it. Because usually what happens is that we build this disconnect, right? When you’re a small clog in a bigger machine, right? When, you know, in enterprise projects, you don’t have the feedback loops from the users, right? As quickly as you would on…
Madhuri Somara (11:40) Right.
Federico Ramallo (11:43) on an internal application as you were sharing before, because there are more steps, there are forms, so you don’t get the excitement until later. So there’s more complexity and delays to get that. So it took me a while to figure out that as well on the product side. I come from a software engineering background.
and I’ve been doing product ownership on some of the projects that I’ve been building. So I’ve been kind of jumping on both sides, right? And it took me a while to understand that as a software engineer, our job is to add value to the user, right? To make sure that we are making their life simple, but basically making sure that what we build, understanding what’s in it for them, you know.
I want to do this and then be able to facilitate that for happening. And when you work on an internal application, it’s amazing because you can uniquely identify the users. They have a name. Whereas an SaaS application or an enterprise project, those names become fuzzy because it’s more of a volume project. ⁓
Madhuri Somara (13:00) said.
Federico Ramallo (13:01) You don’t
know the users by name. You know them by volume, right?
Madhuri Somara (13:05) No, that actually, yeah, those little things, right? So it adds up a lot. Like, you know, if we start enhancing those little things, it actually, it does help customers and end users as well. So as you said, right? So yeah, totally agree.
Federico Ramallo (13:20) Right. So recently you helped launch autonomous agents at Ignite 2024. Can you share the vision behind that project?
Madhuri Somara (13:28) Absolutely, that’s one of my favorite parts of what I have been working on since four years at Microsoft. So the AI agents that I have been working on is case management agent. And obviously, there is…
Definitely an overlap between other AI agents like customer engagement agent or knowledge management agent like these are all powerful AI agents, right? So Let me tell you this So all these agents the vision behind this I like the question, right? So because The vision is what what made possible, you know for building these AI agents. So let me start By saying this it’s not definitely built in silos, right? Because but built very
intentionally to build this, you know, to make this customer experience flawless and you know, pretty smart end to end.
And also, I’m pretty sure it wasn’t built with a mindset like, know, just build it. They’ll come. They’ll come and we’ll get the customers. So it wasn’t built with that mindset. intention and focus was always on solving real customer pain points. And I’m sure you’ll hear me a lot of times talking about customer pain points because I believe in it to hear them out. So, so, yeah, it’s definitely, you know, we are focused on solving those pain points.
when I started working on this, like case management agent which I own, I made it a point to speak to, you know, directly with the end users, like hearing their experiences, you know shadowing them, understanding their workflows, know, where, you know, even their sentiments when they’re dealing with the product, like where are they getting frustrated, for example, right? So the solutions we prioritized truly now they’re in production. It made, it is making their work easier.
Now seeing these external customers use these agents and experience that aha moment which we were talking about, so it’s incredibly rewarding for me, you know, even to see how their lives are becoming easier. So that’s when you know that, you know, the work that you have done truly resonates and delivers impact, right? Truly resonates with end users. So yeah, that’s, I really love being part of it and I’m definitely, you know, thankful to my leaders
chose to be, who chose me to be part of it. So it’s one of the most rewarding experiences for sure.
Federico Ramallo (15:43) Amazing, amazing. Can you expand a little bit on what the product is about, what case studies solves?
Madhuri Somara (15:50) Absolutely.
Absolutely. So I’ll talk about case management agent first. So the case management agent, what it does is,
Every single organization has a support team or a customer service team, right? So they have this protocol that they have to follow from the case creation or an instant creation, even for organizations like Amazon, other huge organizations like laptop companies, like debit card, credit card companies. They do have a protocol or a process that they have to follow from the case creation till the follow up on closure.
So what case management agent does is from the beginning till the end, everything is automated. The case update is automatically done with few very straightforward knobs, where you can set according to your preferences, according to your company. And then how the case is troubleshooted, like how the resolution is found out and you’re sharing with the customer and how you have to follow up and
close the case, right? So all this process is completely automated without human intervention. But I wouldn’t say, you know, that’s one of the things that I don’t want to say it like without human intervention, because we do have human in the loop. Human can always look at like what’s happening on the case, can oversee the case, what’s happening and all of that. So that’s pretty huge, right? Even when I’m saying this right now, from creation to closure, there are like multiple
boilerplate
tasks like you fill the form, you create cases, you follow up, draft emails. There are many things that can be automated, that should have been automated. So that’s what case management agent does.
Federico Ramallo (17:35) Interesting, interesting. Yeah, I think that a better way to phrase it is that the agents can make decisions autonomously or independently and with a high level of trust. Because, you know, I think that’s one of the core issues with this type of automations, right? Can I trust a non-deterministic agent, right?
Madhuri Somara (17:50) Absolutely,
Yep.
Federico Ramallo (17:59) And I think that’s the most important thing. And then instead of human in the loop, because human in the loops means that now the agent depends on the human to proceed. Here you have human supervision, but the agent can independently move forward.
Madhuri Somara (18:15) Absolutely, It’s really, you know, yeah. Now seeing customers use it, like it’s amazing, like what they are saving and, you know, how much time and how much, you know, how fast the resolution is now. It’s pretty amazing.
Federico Ramallo (18:29) Right. Yeah, I’m building an agent to automate hiring and screening for technical talent. And I’m working on the challenges of trust. Can I trust the agent? And how much supervision versus human in the loop do I need to make sure that the quality doesn’t degrade with the automation?
Madhuri Somara (18:36) That’s the reason.
Yeah, absolutely. Let’s be in touch then. I would love to hear more about it.
Federico Ramallo (18:54) Yeah,
happy to share more information about that. Now, what is the biggest challenge when you’re bringing cutting edge AI research with user friendly applications?
Madhuri Somara (18:58) Yeah. Cool.
Right, yeah, like here’s what I think, right? So as a product manager, I’m responsible for designing that entire end-to-end flow, which I just talked about case management agent, for example, And I’m also asking the design and engineering teams to trust my vision and execute on it, right?
So what I think is the biggest challenge in bridging cutting edge AI research with user friendly applications is making sure it truly resonates with the real users. Like, are we solving the real problem or are we just envisioning something which doesn’t make sense at all? So even now when I’m thinking about the product or feature, I always have that thought in the back of my mind. Are we solving the real problem? Is it going to be useful for the users?
So that’s something, you know, because me and my team are investing a lot of energy and wants the solution to genuinely solve customer pain points. So for me, at least, there’s no other way to build these AI solutions than by centering them on, you know, human usability or customer usability and pain points. you know, but.
On the contrary, think that having the tension between technical innovation and usability is actually a good thing. So it ensures that we are building meaningful agents or solutions that truly help people, you know, rather than just being impressive on the paper or rather than just being building something which is not used by the customers. So yeah, I see it as a good thing to have the tension.
Federico Ramallo (20:37) Yeah, I think that part of the process of building a product is about user experience, accessibility. And I used to think that user experience was about make it pretty, right? And that was probably a very short-coming vision from my side, right? But anyway, but then I learned that there’s more to that, that it’s also understanding the behavior patterns.
Madhuri Somara (20:52) Thanks.
Federico Ramallo (21:03) where the buttons should be located, how can you set… Once you have consistency in the interfaces, then the users will be able to find information where they expect it. So building that dynamic, that is the role for the UX. ⁓
Madhuri Somara (21:22) Yeah.
Federico Ramallo (21:25) And I’ve seen how that helps a lot on the product side, right?
Madhuri Somara (21:28) You said it right, the user behavior, right? That’s a key journey that you have to track as well, like how to make it easy for the end users. So yeah, that’s right.
Federico Ramallo (21:39) Right, right. I
evolve, my vision for designer evolved, not vision, but understanding the role that the designers have has on the process of building software products evolved through all the years. Where now I can see how…
bringing a designer from the beginning of getting the requirements and talking to the users can be beneficial and not only make it look pretty, which is the by-product of the UX, but make it user-friendly, predictable, accessible, and an effective tool to use,
Madhuri Somara (22:13) Absolutely, yep, totally agree.
Federico Ramallo (22:15) And then the other part that I learned throughout the process is the UX and the product owners can work together on building more clear requirements for the engineering team instead of having to go back and forth, right? And then I was able to optimize the process. So now the engineering team will not depend as much on the product team.
Madhuri Somara (22:36) Great.
Federico Ramallo (22:36) or
product owners because now they have everything they need to actually be able to build it. And that would make, I mean, as a software engineer, you understand this, right? When you don’t have all the requirements and then you’re like asking questions, what do I have to do next, right? So it’s frustrating on the engineering side, right? So we were able to optimize that and then we were able to…
be able to be more predictable on the deliverables, right? Which help a lot. But then on the process, also learned that on the engineering side, we kind of lose the vision of what’s the value for the user, right? Because we just deal with the product owner, we don’t deal with the user, right? So once I understood that, then I was able to say,
Madhuri Somara (23:02) absolutely.
Federico Ramallo (23:26) what’s the value for the user? How can I build this feature so it can provide the most value for the user? And that changed a lot how I built software after I made that discovery.
Madhuri Somara (23:34) Right.
Absolutely. Yeah, my orange team will definitely agree with that. So, yeah.
Federico Ramallo (23:45) So what role does customer feedback play in shaping AI product experiences?
Madhuri Somara (23:52) That’s a good question because what I think about customer feedback is, for me at least, customer feedback is everything, honestly. So even when I talk about, when I think about the products that I’m building or features that I’m trying to prioritize and launch.
customer feedback is everything to me. As I was also just mentioning about the case management agent, right? So when I started building case management agent, so it was pretty much, I had to make a point that I’m talking to the end users and understand their entire journey, like where I’m seeing that friction, where I’m seeing that, the good part of the feature that they like and what they hate most or what they
you know, what they like most, right? So all of this, you know, for me, definitely customer feedback is really, really key. How I think about, you know, prioritizing features and…
And also, as I said, like I told you, like you’ll hear me saying, you know, that solving that customer pain points is, you know, really important rather than just, you know, thinking about a product feature just so you think it’s fancy, right? You cannot implement something like that. And I really see every interaction with users as an opportunity to understand their workflows, as I was just mentioning. So, yeah, so this, you know,
I think customer feedback is whenever I think about building a feature or whenever I think about anything that I have to build next, I’ll definitely go talk to the customers or go talk to the users to understand what exactly they pinpoint. And even when they’re talking, I sense the sentiment and all of that. this is all.
I think it’s a foundation of building meaningful AI experiences, at least for me personally.
Federico Ramallo (25:41) Right, I agree, completely agree. mean, the customer feedback in any form, it’s the most valuable information for you to be able to execute in your role because you have the power to execute the vision, also, you know, but all start from the input that the customer provides, right? ⁓ Yeah, and there’s a…
Madhuri Somara (26:03) some
Federico Ramallo (26:06) delicate balance of pushing your vision, which requires, I don’t want to say stubbornness, but perseverance, right? Because you’re trying to change, to build something that doesn’t exist, right? And then having the sensitivity to listen what the customer wants, right? And being able to incorporate that into the vision, right? It’s a very…
challenging activity to do.
Madhuri Somara (26:28) Absolutely, right? Yeah, that’s totally true. Like it’s a double-edged sword. Like are you solving the real problems or are you thinking innovatively? It has to be a mix of both, right? So you can’t just stop building something just because, you know, customers didn’t ask for it or, know, so, you know, even with when you’re thinking about a feature, like tying it back to, you know, if tying it back to that pain point, like at least a part of the pain point that if you’re trying to solve.
I think that that’s a delicate balance as you were just mentioning. You can make any feature hit or miss. So you have to find that real ground for sure.
Federico Ramallo (27:05) Yeah. I think this is on, on Steve Shub’s movie, you know, they were saying, he was saying like people don’t know what they want, right. until you show it to them. Right. So I believe that there’s some truth into that, right. but also, I don’t know if you’ve seen the Simpsons when Homer built a car, right. ⁓
Madhuri Somara (27:15) He’s not
that’s.
Sounds good. ⁓
Federico Ramallo (27:24) There’s
an episode where he meets his lost brother. He has a car company. He’s in charge of designing the new car because he’s the common man. The car is a disaster because it incorporates all the crazy features that people ask them without coherence, without thinking about the final product. ⁓
Madhuri Somara (27:41) Yes.
Absolutely.
Federico Ramallo (27:51) what I was talking about, It is balance, right? If you say yes to everything, then you have a crazy product that doesn’t work, right? If you go Steve Shov’s way, then you can get to this minimalist product, right? Finding that balance is a challenge, right?
Madhuri Somara (28:09) I think that’s where the core product management skill comes into handy, right? So you can say no for everything, obviously, and you can say yes for everything as well. So like you have to find that right thing, what matters, really matters, help for adoption as well. You have to think about adoption as well, it’s somewhere in the back of the mind as well, right? So yeah, I think that’s where the core product management skill lies, for sure.
Federico Ramallo (28:38) Right. Yeah. And this applies for AI and non-AI products as well.
Madhuri Somara (28:42) Absolutely, yeah, it’s across. It doesn’t matter, EI or non-EI. absolutely.
Federico Ramallo (28:47) So, changing a little bit of topics, you mentioned in a meeting before about AI evals. So why do you think AI evaluations are becoming so important for product managers?
Madhuri Somara (29:02) Absolutely. that’s another good question. So we have, you know, all of us together, we have entered a new era, like where products don’t just follow instructions, right? So, you know, we used to create workflows, like, know, with giving a set of instructions, like, you know,
If it happens, do this, or if it doesn’t happen, do this. Like that set of instructions were followed by products, but not anymore. Like, you know, we are having these models which are learning and adapting, right? So which means we all have to change together and it’s not just, you know, building those models. have to make them, when the models are going out into customers’ hands, you know, that’s where we have to be more confident, at least for ourselves, right?
you know, these customers, you know, when they are adapting these AI agents, they have to be confident as well that model is acting how they would like to, not vice versa, right? So I think even when I started working on this case management agent, right, like we were not just launching another feature. We were building AI agents that interacts directly with customers on behalf of these large enterprises that we work with, that are our customers, right? That’s a huge responsibility.
on us in intuition or like basic testing, right? It wasn’t enough anymore. It wasn’t enough. I think that’s where the strong AI evals and golden data sets come into the picture. So to be confident of what you’re building and what you’re rolling out to the customers. Like, know, it’s a huge responsibility on our shoulders as well. So that’s where these things are, you know, helping us to make sure we are shipping what customer wants.
we are shipping the models how customer wants technically.
Federico Ramallo (30:46) Right, very interesting, From what you’re describing, AI evals reminds me to BDD on the coding, but here you’re dealing with AI behavior, right? How AI is going to behave and the evals will allow you to make sure that it remains within the expected behavior.
Madhuri Somara (30:48) Yep.
sadly. That’s
Federico Ramallo (31:06) Interesting.
Madhuri Somara (31:07) Yeah, I’m sorry. Yeah. was just saying recently, I started, you know, learning a lot of things about this AI evals and stuff like that. Now I think about it, like we cannot ship anything without, you know, evaluating it, even the, like the models that we are building, right? So it’s just to make sure that you’re building the right product and shipping the right product to the customer. So I cannot think shipping something without AI evals. No. Yeah.
Federico Ramallo (31:07) ⁓
you’re going.
Right.
Madhuri Somara (31:36) Yep.
Federico Ramallo (31:38) It’s when the tools become so useful that now you depend on these new tools to move forward, right?
Madhuri Somara (31:45) Right, yep, that’s true.
Federico Ramallo (31:47) So how do you navigate the tension between AI capabilities and responsible use in enterprise settings?
Madhuri Somara (31:55) I believe we should use AI where it clearly adds value, right? but also set boundaries, you know, just because we can use AI doesn’t mean we should, right?
I go back to this customer pain points and customer feedback, right? when you’re building something, a product or a feature, you know, just because you can use AI or just because you can make a product fancy, we shouldn’t be building those products like that. So you should still ground yourself with, you know, the user experience and, know, with the customer pain points. As I mentioned, like, you’ll hear me a lot talking
about these customer pain points. So absolutely, wherever you see that market fit gap, technically, if you think that product is lacking these enhancements or lacking these capabilities, that’s where you should obviously use AI or you should enhance that product feature.
we should use AI where it really adds value and where it shows that powerful impact for the end users and for the customers. Yep.
Federico Ramallo (32:57) Great.
Madhuri Somara (32:57) That’s how I will navigate
the tension. So I don’t really, I don’t want to go back and forth between, should I use AI or like, should I, you know, it’s just solve the customer pain points. So I will first think about like, you know, how do I solve this issue with or without AI? So that’s how I ground myself, my thoughts, and that’s how I navigate the tension between both of those. ⁓
Federico Ramallo (33:20) Interesting, yeah,
because you’re thinking of AI as another tool in your tool set rather than just AI first kind of thing, right?
Madhuri Somara (33:33) Absolutely, yeah. mean, AI first is something that we all have to agree on. But if there is an opportunity to use AI, definitely. But if it doesn’t make sense to add it there, if it doesn’t make sense to infuse AI there, I wouldn’t really force it. know what I mean? So yeah, that’s what I think.
Federico Ramallo (33:54) Yeah, yeah, and I agree that AI can have so many use cases, it can have a significant impact on product development, both on the building the product, but also adding value to the users. But yes, but I can also see that it’s not a one tool that fits everything, right?
Madhuri Somara (34:12) Absolutely.
Federico Ramallo (34:12) yeah.
And, and that, I think that I’ve seen that that issue happening a lot on the market where, people say, well, I have a hammer. now I, everything I see are nails, right? so, you know, I’ve seen people saying, now that we have AI, I want to apply to everything because everything will be able to be solved with AI, which is not the case, right? Yeah.
Madhuri Somara (34:34) I right? Yeah, yeah, that’s
the vibe that I’m seeing as well. you know, I may or may not agree with it, but that’s exactly the vibe I see it as well.
Federico Ramallo (34:43) yeah, and sometimes it doesn’t make sense, right? It actually overcomplicate things for no reason, right?
Madhuri Somara (34:50) Absolutely, yeah.
one more key takeaway while we are just navigating this tension, right? So we could having that human in the loop, for example, it sounds fancy to just, you know, have no human in the loop. You know, it definitely sounds fancy and that’s what that’s where we are all headed to. But at least for now, I think having that human in the loop or having that, you know, human judgment where it really matters, I think that’s the key as well. Like while you’re developing features.
or while you’re working on the AI features. So that’s something, you know, it’s confidence for the users.
Federico Ramallo (35:24) Yeah, because I mean, for me, automation is great when it works, but when it doesn’t, it’s a nightmare, right? And it can go, yes, and it can go fast very quickly, right? Because now the power of automation is actually multiplying the mistakes that then you have to go back and fix manually, right? So I’m always careful about automating things because of that, yeah.
Madhuri Somara (35:30) Yeah.
Yeah
Lutely, yeah.
Federico Ramallo (35:50) So what advice would you give to women and underrepresented voices looking to break into AI product leadership?
Madhuri Somara (35:58) So, visit!
These are two things that I say to myself regularly. I try to do this as well. So two things, even with my mentees that from my university, like, you know, I coach a few of the university students from my university. I went to Fairleigh Dickinson. So two things that really matters or really, you know, plays a huge role is network actively. Like, you know, you try to build relationships with mentors.
mentors,
peers, and there are tons of resources out there, tons of mentors out there who are ready to coach. You have to do your homework to find them, really good one. So network actively. LinkedIn is the best place to find mentors and other things as well, like peers who can practice with you, peers who can just brainstorm with you and all of that. So networking is really number one for me.
and the second one is curious curiosity like be curious unfortunately this is not something that you know someone can teach you right so you have to be genuinely curious like in this AI world we are all learning together right even now personally I have never seen these many podcasts or you know I have never read these many newsletters in my life along with yours Federico I watch a lot of podcasts and yours as well so there are like
tons of resources out there like be curious you know start you know gathering all those resources listening reading you know that’s the only way we could stay afloat honestly if you ask me and even if you’re working with engineers or designers ask them questions like how did you write this code or how did you build this model and what was your intention with coming up with this design you know those those kind of questions honestly like wherever you see an opportunity
try to learn. That’s what I say to myself and my mentees.
Federico Ramallo (37:48) Right. I, I, my favorite, advice that, that from you is, you know, that they should listen to my podcast. But joking as joking aside, I, I think that’s a great advice. mean, networking, it’s, it’s a great way to, you know, meet people that, in the future are going to,
Either you’re going to be able to help them or they’re going to be able to help you. ⁓ that’s an amazing potential. Yeah. And when is the best time to start? Now. Right.
Madhuri Somara (38:11) Yep.
Now,
right now, totally.
Federico Ramallo (38:20) Yeah.
Yeah. And very interesting that you’re mentoring students. I think that’s amazing.
Madhuri Somara (38:30) Absolutely. Yeah. Yeah. Yeah. I have, you know, a handful of, you know, university students that I just talk to them regularly. I took them to the campus. One of them was here in Atlanta from Jersey. So I took them to the campus and showed him around. And he wrote me back saying it was very helpful for him to put his, you know, career thoughts in a direction. But, yeah, I love, you know, trying to give back what I can. So.
Yeah.
Federico Ramallo (38:56) So talking a bit about the future, where do you see agentic AI and automation having the biggest impact over the next five years or so?
Madhuri Somara (39:09) yeah, that’s a good futuristic question. I think all the…
Federico Ramallo (39:14) Ha ha.
Madhuri Somara (39:16) Over the next five years, think agentic AI and automation, like, I think will reshape everything, you know, how work gets done. not just in customer service because I’m talking about customer service because I do work for a customer service product. Not just in customer service, but across sales, marketing, IT, almost every function, right? So from handling boilerplate tasks end to end, like surfacing insights,
resolving issues autonomously. There’s a ton that it could…
that could happen in the next five years, I think. And with this reshaping, I think this will free humans to focus on most creative, strategic and high judgment tasks, which is important, right? And I’m actually excited to be part of the shift because even now with the copilot, with M365 copilot or researcher agent that I use a lot, I see many of those boilerplate
tasks, like autonomous tasks, I can just give it to Copilot. For example, even if I have to book an appointment for something, I just use Copilot. Compare all these providers or all these saloons and just book something for me. So all these tasks are automated and I could see myself just trusting Copilot, do something for me autonomously while I’m just sitting back and relaxing.
I’m genuinely excited to be part of the shift and you know seeing firsthand how humans and AI together will set the world in motion for sure.
Federico Ramallo (40:48) amazing. ⁓
Madhuri Somara (40:49) Yep.
Federico Ramallo (40:51) So what new skills should future product managers develop to stay relevant in the age of AI?
Madhuri Somara (40:59) Got it. Honestly, I would love to know the answer for that too. But first and foremost, like Satya actually said in his newsletter yesterday, which came out,
Federico Ramallo (41:03) Hahaha! ⁓
Madhuri Somara (41:13) One thing I like what he says always is be a learn it all rather than know it alls like, know You should keep learning, you know, be curious like learn wherever you find an opportunity, right? So I think going back to what we were discussing I would say be proactive and curious Because there are a lot of trends and a lot of things that are happening right now So we can’t just wait for the latest AI trends to land in your lap
So it doesn’t really happen. You have to go after them and we have to seek it out or experiment, learn, read and stuff like that.
That’s that’s and another key takeaway that I have been hearing a lot from product managers these days is you know let’s you’re prioritizing your feature let’s just infuse AI right like again what we were discussing earlier like don’t use AI just because you can even I heard one of our leaders speak the same thing right there has to be a meaningful reason to incorporate something into your products right
AI, I think, should solve real problems and it should not be a fancy add-on, in my opinion. So yeah, those are two things that I would keep in mind as well while I’m navigating this shift along with thousands of product managers.
Federico Ramallo (42:29) Right, right. And I’ve been noticing this trend related to new skills that it’s about because now people are using more and more AI to do their work every day, right? This applies for product managers, but I also apply, but I think also applies for other roles. Yeah.
Madhuri Somara (42:44) for everyone.
Federico Ramallo (42:46) Right, yeah, for everybody else. The thing is, they relate so much that then they lose the ability, they lose the know, know how, how to do things, right? And the underlying whys, right? I mean, I am very familiar with Ruby on Rails. That’s the framework that I love the most. So…
you and probably happens to you with Angular as well, that you get to a point where you know what methods to use to do what you want to do, because you are familiar, you have the familiarity with the framework, right? But then, you know, if I have to work, build a script in Python, I don’t have much experience with it. So I can use AI, just, you know, I wrote this in Ruby, it just helped me translate it to Python, it works, right? And…
But in that process, you don’t learn Python. It used to be that if you need to write Python, you have to learn Python. So I’m finding that this trend would make more valuable the core skills we were talking about before. Understanding how things work to a deeper level.
They said something similar with calculators, right? When kids started to use calculators, you know, they’re going to forget how to do math, which it happened, right? I mean, now kids, well, not kids anymore, I’m grown up, but you know, we depend more on having a calculator to do math, right?
Madhuri Somara (44:27) You’re right, Federico, because when I started working on AI, that’s exactly what I was thinking as well. Probably we’ll lose that basic thinking or basic things that we are doing with or automating with AI now. But I think…
I’m also a firm believer that AI is here to help us and not replace us. Like, you while you said like, you know, we’ll lose the opportunity of learning Python, but we’ll also gain the opportunity of, you know, how to, you know, prompt engineer, like how to play with the prompts, right? So there’s definitely another, you know, direction that we could focus our energy into, right? So definitely, as you said, there is some inevitable loss, right? Like less hires
for manual roles, certain boilerplate tasks disappearing. But that’s part of, as you were mentioning, but that’s part of every major technological shift. So I think, yeah, there’s an opportunity and there’s also some loss, honestly. So, yeah.
We should all focus where we can, focus our energy into different things like prompt engineering or like do something which which matters to you most, right? Like, spending time with families while automating, while monotonous tasks are taken care by AI. So yeah, that’s how I look at it. Yeah.
Federico Ramallo (45:44) Interesting. So we’re running out of time, so I’ll ask you one last question and then we’ll wrap it up. What excites you the most about the future of AI?
Madhuri Somara (45:50) Absolutely.
Absolutely. That’s a… Yeah, what excites me most is how it can take that mental load off so we can focus on what humans can do best, right? Like what we were exactly talking. Like while Microsoft Copilot or, you know, my researcher agent is taking care of…
booking an appointment for me and spending time with my kids. That matters most to me, right? So like these things, like I’m excited to see how it’s taking that mental load off of me, even no shame in admitting it, right? So before AI, I was spending a couple of days, if I have to send an org-wide email, I was spending a couple of days to draft that email, like seeing every single word, proofreading it, like, you know, just drafting it really perfectly. But now…
with AI that’s just doing that, you know, that’s giving that head start for me, like, you know, drafting the skeleton for me so I can just build on top of it, right? So it’s just, now I just probably spend a couple of hours in sending out the org-wide email, right? So that’s how, you know, while it’s automating a few of my tasks, you know, I can do something which is more valuable to me, right? And that…
being said, at the same time, I also stay cautious because AI, are in that midst of like AI isn’t perfect yet. So that’s where we are all headed. Like it’s going to be perfect one day or not. ⁓
Federico Ramallo (47:18) You
Madhuri Somara (47:20) But yeah, I stay cautious. Like there are a few things while it’s booking my appointment, I’ll see like if it’s, you know, using my card in a good way, I have to trust that like there’s a reliability aspect to it, you know, all of that privacy, security, all of these are like really important, right? So I do stay cautious. But yeah, that’s that’s what I think cautious optimism is where I stand today. I think all of us so absolutely.
Federico Ramallo (47:45) Amazing, amazing. I feel we could keep talking for hours, but we’re running out of time. So I want to truly appreciate it for you joining us today, Madhuri. Any last remarks before we wrap it up?
Madhuri Somara (47:50) No.
I really had so much fun, Federico. Thank you so much again for having me here. I had fun. Thank you. Looking forward to talking to you again. Yeah, bye-bye. Bye-bye.
Federico Ramallo (48:05) Thank you. Me too.