[00:00] Pamela Isom: This podcast is for informational purposes only.
[00:26] Personal views and opinions expressed by our podcast guests are their own and not legal advice,
[00:34] neither health, tax, nor professional nor official statements by their organizations.
[00:42] Guest views may not be those of the host.
[00:51] Pamela Isom: Hello and welcome to AI or not, the podcast where business leaders from around the globe share our wisdom and insights that are needed right now to address issues and guide success in your artificial intelligence and your digital transformation journey.
[01:06] I am Pamela Isom and I am your podcast host.
[01:10] We have another one of those special guests with us today that will be Nelli Babayan.
[01:17] She's US Federal AI Director at Microsoft.
[01:21] We collaborated mostly during my tenure while I was at the U.S. Department of Energy,
[01:27] and we have maintained that connection. Hey Nelli, welcome to AI or Not.
[01:33] Nelli Babayan: Hi Pamela. Thank you for having me.
[01:37] Pamela Isom: It's honored to have you here. Will you tell me more about yourself, your career journey and like, how did you end up where you are today? Which is pretty fascinating, of course.
[01:50] Nelli Babayan: So as you mentioned,
[01:51] I'm at Microsoft these days and I'm leading our AI efforts across US Federal customers.
[01:58] I've been at Microsoft for almost six years and it has been an amazing and ever changing journey. That's one of my favorite things of working at Microsoft is that we innovate so much and everything changes at such a fast pace.
[02:11] So you always have to be in that learning and growth mindset and keep up. But my journey to where I am now, where I'm at Microsoft, has definitely not been a linear and this has been a lot of pivots and adapting to various new waves that we're seeing whether,
[02:29] whether it's in industry or anywhere else. I have a very national background when it comes to my career and I've worked in international development,
[02:38] in academia, in policy and think tank analysis.
[02:42] So. And that is where I also then moved to Microsoft. Having having a doctoral degree in international relations with cyber focus, I think prepared me very well to move to Microsoft and then work with our federal customers, know what their mission is, what are the most vital of the government,
[03:00] and to be a trusted partner in that journey.
[03:04] So I think one advice I would give anyone is that be and stay curious about what's going on in your environment and maybe outside of it, color outside of the box and be with the, with the AI development.
[03:17] Be ready to adapt and embrace the changes.
[03:22] Pamela Isom: Well, that's good advice. You have a very interesting background and congratulations on all your successes. I do remember when we first met, I know you were at that time and I look at where you are today.
[03:35] And that's amazing.
[03:37] So let's talk about agentic AI use cases in federal.
[03:41] Can you tell me more about what's going on?
[03:45] Nelli Babayan: Absolutely. We are seeing federal agencies moving towards understanding and adopting agentic AI.
[03:53] And it's not been a new effort. We've been working, our teams have been working with the federal agencies for the past several years,
[04:02] since the general generative AI boom, so to say. And they all started, we all started with some kind of an AI assistant, right. An AI chatbot that we would work with, ask some questions that it answers.
[04:15] But as we are seeing that AI transformation progress,
[04:19] we're seeing also the our customers and including federal agencies move from that first phase of AI adoption of a human with an AI assistant to more agentic use cases and agentic type of workflows within Microsoft and also by looking at our customers globally.
[04:39] We have seen that the AI transformation is domained through this specific journey that we're calling a frontier transformation.
[04:48] And we're seeing that journey happen in distinct three phases. And the phase one is a human with an assistant AI assistant. And that happened also within the federal agencies as they're adopting those AI retrieval agents.
[05:03] Then the next phase we're seeing is really a human with who is joined with various agents that are doing some tests on their behalf. And the phase three where some of our customers are in that phase is really agentic multi agentic system that is doing several tests on whether simultaneously or maybe in a specific sequence set up by a human.
[05:27] And when it comes specifically to federal agencies, we're seeing a lot of diverse type of use cases because we have diverse mission within the federal space. Also think of from case management and information retrieval all the way to more science oriented use cases where for example, some of the labs within the Department of Energy are using AI to improve the efficiency of lithium batteries.
[05:52] So the spread of use cases and the types of use cases is vast. And that is also what's keeping us very busy and very much inspired by how AI can be applied to the mission.
[06:05] Pamela Isom: Are the agents collaborating across agencies?
[06:09] Nelli Babayan: That's probably a great question for the agencies themselves.
[06:13] But we definitely do have some events that we're doing where different agencies can come in and talk about their use cases.
[06:20] Pamela Isom: That's interesting.
[06:22] So let's talk about AI governance and generally AI adoption strategies.
[06:28] I'm really interested in what's working,
[06:32] where are the challenges like what are we doing, what are we getting right and where do we need to really just look at stepping up.
[06:43] Nelli Babayan: I believe what we're doing right is really understanding how we can apply AI purely and effectively. And that comes from the understanding going beyond the hype of AI or going beyond the hype of agentic AI and really looking into the use cases and scenarios and requirements where AI can help with the mission.
[07:05] So we are not adopting AI for the site of AI, but rather with specific use cases in mind that can help with the mission.
[07:14] And this is what we're talking to our customers about and understanding what that use case that you are trying to solve. Right. What are the obstacles that you are trying to overcome, Whether it's in better serving the citizens, maybe reducing time on those case management, or for example, reducing the permitting and licensed bottleneck on nuclear projects.
[07:36] Again, use cases are very diverse.
[07:39] And what we need also need to pay attention to, and I think the customers are doing that and we are helping them with that, is really understanding how to govern those agents.
[07:52] Because we are all used to the usual threat or risks that we have on our environments, whether that's identity. Right. Or a network penetration.
[08:03] But understanding that AI is also bringing additional risks and mitigating those risks.
[08:11] And I always talk to our various customers about those risks and give them the illustration that how they need to address those security issues, if they are. And I'm saying this not to scare anyone off AI, but rather to demonstrate that there are risks.
[08:28] But we also at Microsoft have considered those risks and we have created solutions and tools that can mitigate or even eliminate those.
[08:38] So really, we're thinking about extending to your AI agents the same discipline we applied to humans for decades, like identity, least privilege, continuous verification, observability definitely is a popular one,
[08:56] and the ability to quarantine a compromised agent instantly.
[09:01] So understanding that an agent without a governed identity is an unmanaged insider and there are tools and solutions that we have for that can help with that process as you are moving from your pilots or for proof of concept into production.
[09:18] Because we are going to see agent sprawl from as we are in this process of creating new agents. And we need to be able to govern those.
[09:29] Pamela Isom: So when you talk about AI security,
[09:32] we're discussing AI governance and then we talked about generally AI adoption strategies,
[09:39] and then you started to speak more into the AI security realm, which I agree all of this is interconnected.
[09:47] But then one of the things you said is from a AI security perspective is that we need governed identity of agents.
[09:56] Right? Is that what you said?
[09:58] Nelli Babayan: Yes.
[09:59] Pamela Isom: Can you elaborate on that just a little bit more? Because I agree with you, by the way. I agree that.
[10:05] And the challenge is there are so many identities and how are we going to keep track of these identities? And I just love the expression that you use. It's not just an expression, it's very serious,
[10:21] but I just love it because you said we need governed identity of the agents. Is there more you can share with us on that? Because I do think that I love it.
[10:32] Is there more you can say on that?
[10:34] Nelli Babayan: Sure. It's really about the governance, right? To know what your agent is,
[10:39] where your agent is coming from, who created that agent,
[10:43] what is on the back end of that agent, what platform it has been created with, what is the how it is performing, what data it has access to and which of your users are interacting with that agent and how.
[10:58] So in that way, because you can gain observability into all of your agents that you have within your environment and it doesn't matter which platform they were created with, whether it was a Microsoft platform or something else.
[11:13] But knowing where your agents are, what they're doing,
[11:18] who they're interacting with,
[11:19] that is going to give you that observability to see your entire estate of agents. And this is what I really mean and this is what we have worked towards. Not to go into details or names of the specific solutions,
[11:35] but at your leisure. If you want to take a look at something called Agent365,
[11:41] that is exactly what helps you with the governance of all of your agents across your entire estate.
[11:49] Right. And that also can help with so called shadow AI, been used to shadow it.
[11:55] Now there can be also shadow AI if, let's say your users are trying to create agents with any platforms that may not have been explicitly approved or don't have an ATO within your environment.
[12:07] So that is really what we're talking about when I mentioned governance of agents,
[12:11] observing the agents and securing your agents.
[12:15] Pamela Isom: And we used to think of AI security is about cybersecurity.
[12:19] And I think that what I like about this conversation is we're not saying that it's about cybersecurity. It is, but it's more.
[12:29] Right. And so I think if there's an opportunity, I like to tell my clients, if there's an opportunity to look at how we mature our approaches to AI,
[12:38] we need to look at AI security and understand that there's an element of this that's cybersecurity, but the other element is governance,
[12:47] as you are speaking about here.
[12:49] And so sometimes they'll want to push that capability off to the cybersecurity teams. I'm like, no, no, no. If we're looking to mature as an organization. This is an area where we need to concentrate so the humans are still accountable.
[13:02] Right. And so you really brought that point home. Appreciate that.
[13:06] Nelli Babayan: Yeah. And I think we even discussed it before we discussed data governance.
[13:11] Pamela Isom: Right.
[13:11] Nelli Babayan: And we've talked about data governance for quite some time. Even before the AI, generative AI came.
[13:19] With the generative AI, and with agentic AI, I think data governance is becoming even more important because data needs to be ready before it goes into your AI agents and your AI applications.
[13:31] And very often as we are talking about AI, the questions come up, how do I know that this data specifically can go into this AI agent? And this is when we go back to the data governance conversations of how important it is to have a robust data governance as you are adopting AI
[13:49] Pamela Isom: and it's making the asset management.
[13:52] Well,
[13:53] one would think that this is making the asset management responsibilities more complex.
[14:00] But we could also have agents that assist with that asset management.
[14:04] Nelli Babayan: Absolutely.
[14:05] Pamela Isom: I think to simplify some of these complexities where we have to now start looking at,
[14:10] and you tell me if you agree,
[14:12] we have to start looking at what do we delegate to the agents,
[14:17] because we have a complex environment and we have all these agents, we have all these identities that we have to manage.
[14:25] So now it becomes a question of, okay, yeah, that's true, but that's not a reason to bypass asset management.
[14:31] It's a reason to take advantage of some of these investments, like your investments in AI agents and the agentic programs.
[14:40] And where can we delegate with the proper authorities in place? Because authority is going to always reside with the humans. But what responsibilities can you give to the AI and the agents to assist with this process?
[14:54] Nelli Babayan: Absolutely.
[14:56] Pamela Isom: So now I want to talk about the Genesis Mission. If you can,
[14:59] I'd like you to share what you're doing with that stuff. Like, what's your role?
[15:05] How are you collaborating with the labs if you are in support of the Genesis mission? And not just the DOE in support of the doe, but other agencies, if it's applicable.
[15:17] Tell us more.
[15:18] Nelli Babayan: Yeah, Genesis Mission is an absolutely fascinating initiative and our team at Microsoft is very closely involved. It's not just me who is working on the Genesis mission. We have an entire team that is working tirelessly with the labs and with the Department of Energy on making sure that we are contributing the right way and helping them the mission succeed.
[15:42] So for the listeners, in case they don't know, Genesis Mission is one of the flagship initiatives by the government to have a national scientific platform where you can also use AI to increase scientific turnaround, let's say, or the progress of science.
[16:01] And there are a number of industry partners involved in that mission and Microsoft is one of them.
[16:08] And we have been working with the Genesis mission leads on establishing MOUs and also partnership agreements and also contributing to the various projects that are going to happen within that framework.
[16:24] I can't go into too many details about each of the projects because those are still being decided.
[16:30] But our contribution is also not only with the usual tokens or expertise,
[16:38] but also what we are bringing to the table is our previous work with various national labs within the Department of Energy that been kind of a precursor to the Genesis mission.
[16:51] And I think in those cases the work we have done with the Pacific Northwest National Lab on battery research or with the Idaho National Lab on accelerating nuclear permitting and licensing fits into the framework of the Genes Genesis mission.
[17:09] Among those we are really looking forward to have the national labs work with our Microsoft Discovery team.
[17:17] And Microsoft Discovery is a platform, is an agentic AI platform for scientific discovery, the scientific process.
[17:26] So that entire lifecycle, as you are looking for your research question, then you are generating a hypothesis,
[17:33] you are doing some simulations and experimentation before you do the experiment in,
[17:39] in the real world in the lab, that entire process can be assisted by agentic AI. And that is what our platform is for.
[17:48] And it is going to be one of our main contributions also within that Genesis framework.
[17:55] Pamela Isom: What's it called? Discovery Lab.
[17:57] Nelli Babayan: It's called Microsoft Discovery.
[18:00] Pamela Isom: Okay,
[18:01] so that's interesting insights as to the mission and your roles,
[18:07] you guys roles and what your intended goals are. So that sounds good, I appreciate you sharing that. So I will say as one of the small businesses out here,
[18:18] we are hoping to be able to collaborate and support the Genesis mission as well. And we're doing our part and I'm already doing some work as you know, around microgrids and microgrid discovery and work research.
[18:31] And so ours is AI powered, so ours automatically is linking to the Genesis mission because of the AI advantage that we bring to the table.
[18:39] But it's good to hear what you're doing and some of the ways you're looking to advance that mission. And I'm very excited about more concentration on scientific discovery and the use of AI to help propel scientific discovery.
[18:54] So I like that.
[18:56] Nelli Babayan: It's absolutely fascinating and I think one of the groundbreaking propositions of the Genesis mission is that collaboration between, between the government, public industry and academia and that I think that collaboration that we have already with the doe, with inl, pnnl, other labs and academia across industries demonstrating how we can effectively bring secure scalable AI to solve key energy challenges that we may be having and achieve more of a broader national economic and security goals.
[19:35] Pamela Isom: So from your experiences as AI industry leader and a government contractor,
[19:41] what have you seen and what can you share regarding your experiences from the perspective of successes as well as lessons learned from
[19:53] Nelli Babayan: the lessons learned really the way we are working is to understand the principles that we are aiming to demonstrate publicly and also with our customers is really around security and compliance by design from day one.
[20:07] So security enterprise is built into our solutions and we need to know and our customers as federal agencies I believe know that too. And we security is never is never an afterthought for us.
[20:21] What we also discussing usually is with our customers that we need to understand. Again I'm repeating myself but in the AI adoption understand the use cases, test them, don't be stuck in a POC purgatory but do your pilot or do your proof of concept evaluate before you deploy and then go into production with measurable success criteria.
[20:47] And typically I get also questions around so in that agentic workflow, that agentic framework sounds great, but where the. Where is the human in that process?
[20:57] And I think we need to understand that that human stays always in control or not just in the loop, but it's human in the lead.
[21:05] And where the human is in that process really depends on the scenario that you are solving, on the workflow.
[21:12] These are probably the main lessons learned at the security evaluating before you are deploying. Not being stuck in the POC purgatory Keep humans in control.
[21:22] Pamela Isom: So what has surprised you when it comes to AI and particularly AI and federal or AI and energy. Have there been Tell us what surprised you. It could be good or bad or both.
[21:34] Nelli Babayan: I think those those were not complete surprises but actually surprises that we have been working towards.
[21:41] And it's always pleasant when your goals come to fruition. Right. I think one of the from different work projects that we have done, I think sometimes the compression of the timelines of how much AI can help with efficiency and productivity.
[21:58] And for that I can bring the example of the work we have been doing on reducing the nuclear permitting licensing debutments.
[22:07] As you know it takes a lot of time and years, hundreds of millions of dollars and immense data processing that may go into that process of permitting and licensing for nuclear projects.
[22:21] And we have been working on a solution that has also been adopted by the Idaho National Lab story to compress those timelines.
[22:31] And we have seen tremendous success in that case, we have partners and customers who have reduced their timelines by 70, 80%.
[22:39] So that's a huge progress and also a very pleasant surprise. Right.
[22:45] But I think it's also the dual role of AI in energy because AI is simultaneously a new energy consumer,
[22:55] but also one of the most powerful tools that we have to accelerate reliable energy,
[23:02] just as I described with the nuclear permitting.
[23:05] So yeah, again, think about the design, permitting, grid stability you mentioned and predictive maintenance.
[23:11] So, and we have collaborated with other partners such as Nvidia, for example, on bringing AI into energy and accelerating those timelines.
[23:20] And so I would say probably what surprised me really with those use cases that I mentioned is the symmetry. AI is kind of power hungry for power, but at the same time it's the best tool we have to bring safe and reliable power online first.
[23:37] So we're permitting to maybe several months before it can be done in few days.
[23:42] So that's not just an efficiency story, but it can also be a trust story where every decision going back to our conversation about security can be traceable and can be audit ready.
[23:53] Pamela Isom: That makes a lot of sense.
[23:56] I give the guests an opportunity to share anything else that they want to share with us and then close out with words of wisdom or a call to action or both that you want to leave with us.
[24:10] So take it away.
[24:13] Nelli Babayan: I think I mentioned in the beginning that I love my role and my work at Microsoft because it's so fast paced.
[24:22] And I think my words of wisdom could be move fast,
[24:28] but at the same time be accountable of what you're doing and how you are also adopting AI.
[24:37] Because as we're looking at the organizations that are adopting AI, they're winning with AI.
[24:42] Build guardrails into the same motion as their rollout because they don't sacrifice quality, they don't sacrifice accountability or intelligence to move fast.
[24:53] But at the same time, as they are moving fast, they are not being left behind.
[24:57] So, and I think my call to action would be for any organization,
[25:02] including our federal space, is to start experimenting with AI and in their accredited environments, with a governed evaluated agents in a mission work in a workflow that can contribute to the mission.
[25:20] Pamela Isom: Yep, I appreciate that. That's awesome.
[25:23] So I want to thank you for taking time to talk to me today and being prepared to share your insights with the audience.
[25:36] So I really appreciate it and it's good to see you. It's been a minute, but it's good to see you. So thank you for being here.
[25:42] Nelli Babayan: Thank you, Pamela. And let's not wait another minute, all right.
