Ep. 17: Copilot AI and the Steps to Transform Your Enterprise

Summary

AI – adopting it and adapting your enterprise to it – is at the top of nearly every organization’s agenda. And yet while 79% of leaders agree with that priority, more than half fear they lack the vision to pull it off, according to a recent survey by Microsoft and LinkedIn. This is where experts like Dona Sarkar come in. She directs Microsoft’s Copilot and AI extensibility program, teaching enterprise leaders how to incorporate this game-changing technology effectively and seamlessly, all while staying on the lookout for AI bias and other pitfalls.

Bottom line: Play an active role. “Happen to AI,” she says, “rather than letting AI happen to [you.]”

Host: Doug Thompson, chief education architect, Tanium
Guest: Dona Sarkar, chief troublemaker, Copilot and AI extensibility program, Microsoft

Show notes

For more info on the best ways to integrate AI into your organization – and to get the most out of Microsoft Copilot, in particular – check out our articles in Focal Point, Tanium’s award-winning online cyber news magazine, and these other useful resources.

• 4 Critical Leadership Priorities in the AI Era | Focal Point
• AI vs. Humans: Why SecOps May (Not) Be the Next Battleground | Focal Point
• Preventing AI Bias Starts at the Top – Just Ask These Female Chief AI Officers | Focal Point
• Tanium Integrates with Microsoft Copilot for Security – Changing the Game for Cybersecurity Teams | Tanium
• Data Sheet: Tanium & Microsoft Copilot for Security | Tanium
• On-Demand Webinar: Empowering Microsoft Copilot for Security with Real-Time Data | Tanium

Transcript

The following interview has been edited for clarity.

Companies really do not have their data in order. And one of the things everyone says is, oh, AI has made us so much more vulnerable with data issues. I’m like, nope, AI is a big flashlight that’s shining a light under your bed and saying mothballs. And you’re saying, oh, where did those moth balls come from? Those have already been there, my friend. It’s just [that] AI’s putting a big spotlight on it.

Doug Thompson: Tech disruption isn’t just about introducing something new. It’s also about gaining people’s trust to transform behavior, business, and the bottom line. That’s what we’re seeing today as enterprises try to adopt AI. Tech disruption isn’t easy. And here are two stats to prove that: A recent survey from Microsoft and LinkedIn found that 79% of leaders know they must adopt AI to stay competitive, but 60% worry their organization lacks the vision to pull that off. We saw this last year with ChatGPT and this year it’s Microsoft Copilot.

Enterprises want to adopt AI but aren’t sure how to get the most out of it.

Hi, I am Doug Thompson, chief education architect here at Tanium, and today on Let’s Converge, we’re talking Copilot AI and the tricks to unlock its full potential.

Joining me today is Dona Sarkar, who is the ultimate shapeshifter. At Microsoft, she’s worked as a software engineer, a product manager, the chief #ninjacat for the Windows Insider program, where she oversaw 21 million users who helped co-create products. She served as director of technology for accessibility services and today chief troublemaker, directing the Copilot and AI extensibility program. She’s also an author, a fashion designer, and – my personal favorite – co-owner of a wine bar. No wonder she was named one of Fast Company‘s 100 most productive people.

Welcome to the podcast, Dona.

Dona Sarkar: Thank you so much, Doug. It’s so fun to join you. And thank you so much for that fabulous introduction because I would just like you to follow me around and introduce me. That sounds very impressive.

Thompson: I am available for those things that. But I have to ask this question: Are there more than 24 hours in your day? Seriously, how do you manage to get so much done?

Sarkar: It’s one of those things where I have given up on the idea of getting it all done, and I’ve really embraced this theory of getting exactly two things done a day. And that’s it. So pretty much a day like today, where I’ve been doing my day job and then in the evening I’ve got some wine bar stuff to do, and that’s it. That is all I’m going to get done today. Yesterday, for example, I needed social time with friends. So I did my day job and had social time with friends. So I book myself for two things a day and that’s it, that’s it, that’s it.

Thompson: And by doing this personal, side-hustle stuff, as we call it, you’re sort of fulfilling all your life goals. We have one life to live.

Sarkar: Yeah, it’s so true. One of my things is that: What are some things that I would want to do in retirement and how can I start doing them now as a side hustle? So then when I do retire – air quotes “retire” – from tech, I will second act into them, but it’ll be something I’ve been doing for, I don’t know, 10, 20 years. So I’ll already be kind of good at it.

So I am a huge fan of future-proofing by having a bunch of side hustles going that you can pick up as a full-time thing while you seek your next job. If you want to seek a next job at all, or you want to take one of your side hustles and now run it as a full-time thing.

Thompson: That’s outstanding. Well, I rattled off all the bullet points in your resume, including fashion diva and newbie bar owner. You’ve worked with customers with disabilities, you’ve worked with Windows Insiders Group, you’ve also worked with Microsoft Power Apps, a suite of low code services to build business applications. That’s a vast array of customers. How have all these experiences sort of impacted the way you see Copilot and AI in general?

Sarkar: So one of my themes for work, and it’s always been true, I’ve now been in the industry since 2002, so oh my gosh, that’s 22 years. That’s a long time. So I’ve been working with customers full-time as a dev, always customer focused, always customer focused, now for 22 years. And my big theme has been tech democratization. So it’s how do we take a tech piece of technology and make it something that is relevant towards the largest audience possible? And when I talk about tech democratization, people think, oh, you mean this industry or that industry? No, I mean globally. How can the tech we build be relevant in Lagos, Nigeria, or Uruguay in South America, or something relevant in Jakarta, Indonesia, as well as Birmingham, Alabama. Because one of the things that we know we have in the world is digital inequity – because not everyone has the same access to internet knowledge, role models, opportunities, et cetera, et cetera.

The tech that I tend to work on is usually something that is very focused on tech democratization, like Windows Insider program. We have insiders in every country in the world. And the point of it is to give anyone who wants to access to an earlier version of Windows so they can skill themselves up and their organization before it goes out to a global rollout so they can be better prepared. And the same thing for the Power [Apps] platform, low code, right? I’m a high code person, but I realize not everyone in the world has time to go get a computer science degree or do a coding bootcamp. So Power Apps gives you the ability to build enterprise grade apps for your business with drag and drop a few formulas, et cetera. And the same thing for accessibility, making sure everyone in the world who has physical disability, neurodiversity, or mental health condition is able to use technology to live a freer life.

After I worked with all of these different kinds of groups of people with all this technology, I realized AI is actually a really good equalizer if we let it be so. It really does become an amazing tech mentor for people who may come from non-technical background, or it might be a great translation tool because people might be working on something that’s not their native language. And it can be something where it allows them access to things that people may not be able to afford, like legal counsel or a preliminary medical examination or that kind of thing. So I think AI has the power to be a very good democratization tool, and that’s what I’m most excited about, completely out of this whole entire AI-verse that we’re living in.

Thompson: That’s wonderful. I really like that. And y’know, I’ve followed your career for a while. We worked not together, but at Microsoft at the same time. What are some of the biggest misconceptions about AI and Copilot in particular? I mean, you mentioned some of this democratization and stuff like that. Certainly there’s some themes that come up.

Sarkar: Yeah, I think one of the biggest misconceptions is that AI and Copilot will make us lazier. Well, actually, there’s three misconceptions that let’s talk about. And I hear this in every country I go to.

The first one is that the use cases for AI is everything or it’s nothing. Either a business is going to completely go in on AI or they’re not. So that’s the first one. The second one is that every business is ready to AI. It’s: “Okay, our data’s ready, we’re ready to roll tomorrow.” And the third one is that everyone knows how to use it. None of these things are true at all, at all, at all.

So the first one: Every company thinks, oh, we need to apply AI to every single scenario. And that’s just not true. Because generative AI, the thing we’re all talking about now, generates. So it’s really good for use cases where you get a non-deterministic output – so where the output does not need to be an exact retrieval of some phrase.

AI is not search. And when people use it as that, I’m like, you CAN, but the real power of generative AI is generating new things. Generate me an email that schedules time with Doug to do a creative interview; create me a strategy paper; create this new picture, a new video. So create from scratch new, where creativity is actually valued. I always say about AI, think of it like you think of the internet: It’s something that can help you solve new problems using new ideas. So I use AI in scenarios where it’s celebrated and not just tolerated, right? Rather than every single scenario on the planet. So that’s the first one.

Second one is: Companies really do not have their data in order. And one of the things everyone says is, oh, AI has made us so much more vulnerable with data issues. I’m like, Nope, AI is a big flashlight that’s shining a light under your bed and saying mothballs. And you’re saying, oh, where did those mothballs come from? Those have already been there, my friend. It’s just [that] AI’s putting a big spotlight on it. So that security by obscurity, where it’s hard for people to find, no longer applies. We actually do need to get our data security in order before we can implement AI. There’s a lot of tactics to doing this, but I think that has made people realize like, oh, actually we weren’t as good a data security as we thought.

And the third one is: People will figure out how to use it. And that’s just not fundamentally true because it’s not that natural of a thing. People have not figured out how to use anything without proper training, without guidance, without saying, “This is a good use case; here’s a good prompt; use it for specifically this thing.” That is what people actually need, and companies absolutely need to invest in skilling for their organizations in a way that makes it integrated into their job. Rather than saying, oh, my finance team needs to learn it on the weekend. Why would they do that? They have things to do. So this is where I found hackathons, prompt-a-thons – one prompt a day training – that kind of thing, to be really, really good and for this to work very well in organizations.

Thompson: That last point, the organizational change or the culture change that needs to take place is very important, and people overlook that a lot. They think it’s a technological problem and technology is fine, but it is a cultural and organizational thing as much as it is anything else.

Sarkar: Yes, and that’s very true. AI is about 50% people and 50% tech. And getting your tech in order so you can do AI is great. Defining the use cases, making sure your data’s in order, making sure that you’ve implemented the right AI product – that is 50% of the work. The other 50% is teaching your org to use it in a way that makes sense so they feel empowered and not scared. That is key.

Because, yes, AI is going to automate many parts of jobs. AI is not going to come and take your job. It’s going to automate away tasks in your job. Like me as a developer, yes, AI is going to probably build the first version of every website I make from now until eternity. That is great news. But I as the dev’ will need to do the prompting to say, build me a website for a wine bar in Washington state with natural wines pulled from these five websites. Have the theme be turquoise and gold.

Thompson: As a writer or content creator or just keynote speaker stuff, I use AI to do my first draft, like you said, because that saves a lot of time. Then I go through it and I personalize it and I say, yeah, this is a good thing or not, I go research, things like that. So it definitely has, like building that website, it definitely saves me a lot of time to do that.

Sarkar: Exactly.

Thompson: So let’s shift gears a little bit here. I wanted to ask you about AI bias, and this is a huge issue and in many ways it’s almost seems insurmountable. It stops business leaders in their tracks. They want to implement AI, but they don’t want to feed it biased data. How do you avoid bias? And this – probably your past history and some of your other roles probably helps here a lot.

Sarkar: Yeah. Bias is really interesting topic for me because it all comes down to what do we consider bias? Because what I consider bias and what you consider bias will be really different, even if we do the exact same job. So a lot of it comes from me asking the question, what is this for? So, for example, I’m a huge believer in putting guardrails around AI to only do the thing that AI tool is supposed to do.

I’ll give you an example. Say you’re building a tool for doctors to – I just saw it, it’s called Color [Health’s] “copilot” and it helps cancer detection in patients. So let’s just eliminate some basic bias. If someone comes to the copilot and says, “Hey, help me buy a Ferrari,” the copilot should say, “I don’t do that. I just help with cancer detection.” That’s all. So one thing is really thinking through what can the AI product do or not do?
And that’s just a matter of specifying the system prompt of the API call and saying, all you do is answer cancer questions. You don’t answer anything else. If someone asks you anything else, give this message. So that’s step one for responsible AI – only answer questions about this.

Second thing is looking at what data is it ground on? Because this is where my job comes into play with extensibility, right? Where’s the data coming from? Is it just the standard native GPT-4-Omni or is there some sort of a dataset that comes from this cancer clinic’s patient records or research or something? I assume it’s ground in some sort of data.

So now let’s think about the data. Do they have an overabundance of data from people who live on the West Coast of the U.S. of a certain income bracket, or do they have data that is pretty representative of the global audience? Pretty sure they’re going to have skewed data.

But this is where the third part comes in, which is how do you equalize the bias in AI so that the abundance of data doesn’t always take precedence? So say they upload a DNA strand, they’re like, yep, this person is likely to get cancer, if they’re of this demographic, et cetera, et cetera, et cetera. So this is where this concept of fine-tuning with data, or retrieval-augmented generation [a process of a large language model referring to an independent knowledge base outside its training data to improve its response], or something with changing the weights of data so that the data set is not as biased as the training data is.

And this is where we’re going to spend the rest of our careers, Doug, you know this. [He laughs.] Where we’re going to be saying, where are underrepresented people in this? And it can be language, it can be race, gender, location, age – age is a big thing. And how can we make sure that people are getting equity in treatment based on the fact that we just don’t have equal amounts of data for equal amounts of people? It’s a very complicated question, but the short answer is we all have to get more technical about how AI works, so we’re able to implement as many de-biasing techniques as possible to solve a very specific problem.

Thompson: I was talking to somebody else in healthcare and AI and she gave a similar example to that of treatment plans that were overloaded with older white males, that they’re just over-represented. Thinking about where that data comes from and how to sanitize it and make sure, again, everybody’s represented.

Sarkar: That’s right.

Thompson: What excites you most about AI? Is there some benefit you didn’t expect, something you didn’t see coming into this role?

Sarkar: One of the things that to me is so interesting is having this permanent thought partner for any topic in the world. So, for example, I was making a negotiation tutor the other day. I said, help me be better at negotiating. Negotiating what? Anything. The price of this rental car; more weeks of vacation; where to go for dinner with my family. So it really helped me realize, okay, I’m very good at this part of negotiation. I’m not very good at this [other] part of negotiation. And I just set up a basic negotiation mentor using Microsoft Copilot and said, “You will be my negotiation mentor. Help me negotiate various scenarios. You give me a problem, I’m going to come to you with a solution. Once I come to you with a solution, give me feedback on my solution and help me create the next negotiation chip.” It’s just basic things like this.

We never had this before. We’ve never had this before, and this is something I can just use in the privacy of my own Copilot so I can learn to be a better communicator, a better listener, a better negotiator. So my next one will be help me be a better listener, help me be a better leader, help me be a better manager, because I think this is one of the greatest mentoring tools that has ever existed.

Thompson: Yeah, you’re constantly developing that. Over the years I’ve known you, trying to develop those skills is something you’ve always focused on. My wife caught me the other night arguing with my ChatGPT over something because it wasn’t understanding my prompt. And I simply wasn’t telling it what I really wanted to get out of it. And that’s another thing, a future career thing, for people listening. Prompt engineering, that’s going to be huge.

Sarkar: Yeah. And prompt engineering will be part of every job. It won’t be so much like a job title, but it will be something that is expected from jobs, where people who are either entering the workforce or pivoting their career or whatever, they will be expected to know really good prompt techniques for their specific industry to solve specific problems. And that’s why I think it’s really important for people to, again, get more technical, get more hands-on and say, okay, AI is really good at these things, less good at those things. So anyone who wants to be a leader in their industry, they actually should start creating a prompt library of, here are 10 great prompts for lawyers who work in accessibility, or here are 10 scenarios that are really, really good for AI in cancer research, and here are 10 scenarios [that] you should not use AI in cancer research. I think being able to also differentiate and say, use AI for this, not for that, that’ll really showcase people as being thought leaders in the space around their specific industry and their specific use cases.

Thompson: I agree with that. It’s part of everybody’s job. It’s sort of like being able to use an office suite or something. That’s something you just do every day. So if there’s one thing, as we get close to the end here, if there’s one thing business leaders should take away from this conversation, one thing that they need to keep in mind about how AI can transform their business, what would that be?

Sarkar: I think the main thing is every business leader has to know that AI is going to transform their business in some shape or another. It is going to happen at some point. It may not happen tomorrow morning, but it will happen in the next few years. It’s like when the internet became common, everyone’s like, oh, we’re not going to have our company use the internet, we’re just going to all be internal only. That lasted like a year and a half, right? It just doesn’t make any sense. The same thing with mobile. Every company has a website that works on mobile.

It’s just going to happen. The best thing they can do is play an active role. Go and happen to AI rather than letting AI happen to them.

Thompson: No, I think that’s right. You can’t hide your head in the sand on this one. This is something that you definitely have to consider. But then again, you can’t just jump in and buy anything because it’s been powered by AI or something. I mean, that’s just as foolish as trying to hide from it.

Sarkar: That’s right. That’s why it’s really important to be intentional and say, which of these processes do we want to actually use AI to improve? And how can I scale up my organization in using it to do things?

I would say it’s a three-step process. One, get your hands on [it], with just free AI tools that exist today, like Copilot or ChatGPT, whatever. See what AI’s good at and what it’s bad at specifically for your industry. Second, figure out what tool is best for your organization. It can be a targeted tool. It could be a large-scale tool. And three, figure out how you’re going to get your data in order and scale up your organization to actually be able to use it to solve very specific scenarios in your business around your space.

Thompson: Yeah, and do that third one in parallel with the other two. Don’t wait for the last part of it.

Sarkar: No. Definitely.

Thompson: So Dona, thanks a lot. It is always an exciting conversation when I get to talk to you. Thanks for sharing your newest role and passion. I followed you all through Europe in your last [project and now] Copilot. It’s just phenomenal, the people that you’re impacting and the things that you’re doing, so I really appreciate that.

Sarkar: I love it. I love it. Thank you so much for having me on, Doug. Thank you. Thanks everyone.

Thompson: I’ve been talking with Dona Sarkar, chief troublemaker at Microsoft.

If you’d like to learn more about how AI can transform your enterprise, check out Focal Point, Tanium’s online cyber news magazine. We’ve got links to several articles in the show notes. Or visit tanium.com/p for publications.

To hear more conversations with today’s top business leaders and security experts, make sure to subscribe to Let’s Converge on your favorite podcast app, and if you like this episode, please give us a five-star rating.

Thanks for listening. We look forward to sharing more cyber insights on the next episode of Let’s Converge.