Making AI Deterministic: Outcomes, Trust and Scaling Enterprise AI

Victoria Mensch: Hello and welcome to our series Decisions at the Frontier, run by Silicon Valley Executive Academy. Today our guest is Harish Epuri. He's a VP of engineering at Autodesk, and he's also a technical advisor to startups in the machine learning and AI space. So we're very excited about having Harish with us. Welcome. What is the most pressure, from a decision-making point of view, that you're feeling right now?

Harish Epuri: There is a lot of potential. How everyone can take advantage of this in a right way is what keeps me awake, and most of the other leaders are also thinking about it.

Victoria Mensch: So the question is, what would be the right way, Harish? What are the challenges, what are the concerns, and what would be the right way in your opinion?

Harish Epuri: I feel personally this GenAI is as powerful as the internet, or more than that. It is impacting not just software but every other industry. What I personally do is go back to the basics. Technology is technology. You've got to evaluate it and use it for the right reasons in the right way. Those fundamentals, to me, do not change, and if you do that, you can embrace any technology and make best use of it. What I've seen some people doing is trying to force-use it. That's where you start getting into mis-expectations, and then you've got to manage all that misalignment. So go back to fundamentals: use the technology where it matters. If it meets the purpose, take advantage of it. First you've got to assess your current organization, where the teams are spending time. Start slow, start in one area where you think the teams are spending a lot of time, and if you see a good opportunity where AI can help, then go and start using it. To me it's taking baby steps, and at every step measuring the outcomes: are we moving in the right direction? There is a lot of talk about metrics like velocity, sprint, time to market. Those are great, but ultimately you need to focus on the outcome. If it is not outcome-driven, all these metrics are meaningless, because we can satisfy a lot of these metrics, but if the outcome is not achieved, this becomes another technology failure. I have seen this multiple times. So outcome-driven is very, very important. The other metrics will help, but they should map to the outcome.

Victoria Mensch: One of the conversations I'm hearing a lot is the difference between an increase in productivity and transformation of the process. What is your view on that?

Harish Epuri: We have done a lot of automation. People have tried RPA, people have done scaling; a lot of techniques have been used. But the fundamental difference in automation is that you are defining a finite set of activities. You have programmed it, you know what the outcome is. Now with generative AI you are asking a machine to do it, and you are also asking the machine to take the corrective action. Productivity matters, time to market matters, but you need that human oversight to make sure the outcome is what you are expecting and there is no drift. At the end of the day, the fundamental difference is that automation is very deterministic. What AI is doing is probabilistic. You need to reduce the gap and go more in that finite direction where it becomes deterministic, so that no matter how you give the input, the output is what you are expecting.

Victoria Mensch: And how would you measure outcome?

Harish Epuri: An example: if you are using artificial intelligence to perform a lot of data calculations and number crunching at scale, the outcome is, am I producing the same result in a much smaller time frame, are the numbers close, 80 to 90%, and what human oversight do I need to put in to make sure the numbers are matching? Once your numbers match what you were getting with the previous process, how do you take advantage of more advancements in the technology and get better insights into that data? As a human, in an existing process, maybe you had finite dimensions of how you were looking at it, but with AI your possibilities are increasing, so you can have more insights, and more insights could help you unlock more value from the same data. That's an example of how you can take advantage of the technology and increase your outcome.

Victoria Mensch: So from the adoption perspective, where is that bottleneck, do you think?

Harish Epuri: In any technology adoption there is a fear of change: when things are working, why change? That's a human mindset, and that's where leadership comes into the picture, to say why you want to change. You've got to put forward the right reasons. Then the challenge leaders face is how to get buy-in from team members to adopt this thing. What's in it for the company, and what's in it for them? At the end of the day, teams also need to feel: if I do this, how is it affecting my career? Am I going to enjoy it, am I going to learn something new, will it advance my career? So you define that and link the whole work. Are they using the technology the right way? If they don't know, how do you enable them, how do you get them the right training? Or you can get the right external help, because as technology is changing, some of your internal folks may not know how to use it. Bring in the right people to help them out. Once you enable them, these are smart people; they'll figure out how to do it.

Victoria Mensch: So how do you build that trust with your team so that they would actually follow the change agenda?

Harish Epuri: Trust is not something you build overnight. When you are forming teams, you are hopefully hiring the right people with the right mindset. When I hire people I look at multiple dimensions. The skills are basic. Second, how curious are they? Can they handle the pressure? Can they work in a team environment? It's not everything about me; at the end of the day you have to succeed as a team. When you look at all these dimensions and get people who are eager to learn, the whole environment becomes different. The best way to build trust, I would say, is you take a commitment and put your best effort into achieving it. What's important for me is the effort they are putting in, not the result, because the result is sometimes not in anybody's hands, but what's in your hands is effort. Once you do that, people realize it. You don't need to go and justify to other people that you worked hard. That trust automatically establishes, because the other person sees that you are really putting in the effort, and if the result is not coming, there are many people willing to help and correct the course to achieve that outcome. That, to me, is how you establish trust. And then there are other techniques: be transparent. People make mistakes; give them the space to operate, and create an environment where risk-taking is possible. As a leader, make sure the team is protected. There is a pretty good phrase: success is all yours, failure is mine. That creates an environment where people can thrive. That's how you build trust with the teams.

Victoria Mensch: You led big technology initiatives within organizations on the engineering side. One of the things that consistently comes up when we talk to executives across industries, countries and regions is change and transformation. How do I actually go about change and transformation without breaking the core? Why should I change if everything is working? That's a big topic not only on the individual level but on the organizational level, especially for established companies that already have a core business that's working. How do they go about transformation, new projects, new products and new technologies without risking what's working?

Harish Epuri: One of my learnings from past experience is that a lot of times transformations fail not because of technology. I'm sure there are studies by Gartner and others: these transformations fail because of a lack of alignment and collaboration. If you do not have alignment on the core objective, on why the transformation is needed, and then collaboration as a team, it doesn't work, because these transformations are not in a single team. They are across the company, across multiple departments: marketing, finance, technology, product. A lot of groups need to come together. Everybody needs to understand their role. Everybody needs to understand who the decision maker is in each area. People can always have inputs in other areas; how do you create space for those inputs, but still go with the mindset that you are only giving an input and ultimately the decision belongs in that area? Those are some of the foundational principles I have observed and learned, and this is not a solved problem; otherwise all of these projects would have been successful. I always try to do my part, and if everybody tries their part, hopefully success will come. Fortunately, I have been lucky to work on projects where that collaboration was there. Sometimes I was able to influence; sometimes I was able to go with open ears and listen to the other perspective. That's also very important. Sometimes you think, I know everything on technology, let me make my decision. Sometimes you just step back and say, this person is saying something, let me listen to them first before I respond.

Victoria Mensch: There is an almost inherent tension, especially in high-tech organizations, and I'm sure in other industries as well, between the technology or product side and the business side. These are two different points of view. I don't want to say they collide, but there's definitely tension, sometimes healthy, sometimes not. I know you have a strong point of view on prioritization and stakeholder transparency. What's a decision you made, or saw made, that changed the conversation from a purely technical activity to business impact?

Harish Epuri: That's such a great question. We live and breathe this every day. I work with the marketing department, and I worked in fintech companies. Marketing organizations used to be mostly crunching numbers, running reports, running campaigns, but in the last 15 to 20 years the technology has changed so much. So many vendors came into this space, and rather than pitching to an IT organization, most of these vendors pitch to the marketing organization. A lot of times I get requests saying, can you implement this, and I'm scratching my head: what is this, give me some context. You can't fight against the tide when the whole shift is happening. What you need to do is say: yes, I'll implement this, but tell me, what does success look like? Just ask that question rather than saying no. What will this achieve? That starts the conversation. Once the conversation has started, you can ask more. An example: I've been told multiple times, we need a headless content management system. And I said, okay, you already have a content management system; we can make it headless. What do you want to do with it? Then you ask certain questions: how is the content managed? Do you want AI to generate the content? Is there human oversight? If the content is not right, who takes the call? Who is on call to handle this? Your website and brand are at stake. If you ask all these questions, that initial friction point becomes a healthy conversation where you are trying to solve a problem, rather than feeling you're being told to just adopt a technology.

Victoria Mensch: Did you have a situation where somebody came with a request, let's implement this feature or that feature, and you had to say no? How did that go? How did you make that decision?

Harish Epuri: I'll be very candid: I made many mistakes. I said no for the right reasons, my intent was right, but it was misconstrued. Sometimes it could be priority, sometimes lack of bandwidth, sometimes it's not the right fit. You've got to work with the right level of stakeholders. Instead of a plain no: yes, I want to do this, but here are the things I'm actively doing; which one do you want me to stop? The question goes back to them. That's one way to do it. Or: which of these can take a small delay? Or: there's already an activity we are doing that's very similar in nature; why do you want to redo it? Those conversations definitely help. Sometimes a no is taken as the person not being willing to collaborate, but if you explain the why behind the no, it really helps.

Victoria Mensch: What leadership challenge has become unavoidable in the last 12 to 18 months?

Harish Epuri: When GenAI came, two, two and a half years ago, the first couple of months people were very excited about it. Then a lot of companies saw the potential, and in every meeting at every leadership level there was only one topic: what can we do to leverage GenAI? I have seen many projects, and talked to a lot of colleagues and friends, where projects were forcefully done because people were in the mindset that we are going to miss the boat, let's adopt this. Of course, coming back to my outcome-based point, there was no proper due diligence and not a lot of alignment, so those things will not work out. I think companies have realized it. From late last year to now, in 2026, people are really using AI where it makes sense. Velocity and productivity, of course: use it in those places where it can solve a portion of the area, but don't overextend it. And the last thing: at the end of the day you all want to do a great job, so have that human intelligence, or human governance, because GenAI can give a lot of hallucinations. There need to be controls to make sure things are working fine. If you put a chatbot on a website, which pretty much every company does, not every question is tested; people are asking lots of questions. So you have to put a framework on the technology so it operates within those guidelines, and if it goes outside that framework, it automatically alerts you: something came out of this boundary, let's have an agent or a person handle it. That's very important, because the last thing you want is to misguide a customer or give wrong information. So to come back to your question, companies and leaders have started to slow down, take a look at what's on the table and how fast it is changing, and they are trying to use it in the right way.

Victoria Mensch: So getting from the experimentation or pilot phase into more of a scaling and growth phase with this technology.

Harish Epuri: Yes.

Victoria Mensch: And what are the main challenges in moving from the pilot phase into the scaling effort?

Harish Epuri: It's a great question. I get pinged by many vendors saying, we've got a great technology. The demo use case looks great, but when you're faced with large volumes of data and a diverse set of data, does it handle all the use cases? That's where the scale challenges come in, and that's where domain expertise is very important. People with deep expertise need to look at it. Those are the scale challenges you cannot assess within a one- or two-hour demo of any tool. That's where the challenges come in, from the pilot to the real implementation. An example: you can put up a website very well, but now you need to do it worldwide in many locales and languages. Suddenly you are talking about how you handle all of that. If you choose the right vendor or partner, they will also work with you on their roadmap to address those shortcomings, and that way it is a win-win for both teams.

Victoria Mensch: Thank you, Harish. Let me switch gears a little and talk more about leadership style, on a personal note. How has your view of leadership evolved over time, given all those years of experience?

Harish Epuri: Every day is a new learning. Every day you face challenges. This is where servant leadership comes in. As a leader you always carry a huge responsibility, but you don't go in every day thinking, I'm in charge. You go into the office asking: how can we achieve our outcomes, and how can I enable my team? It's not that you are in charge. It's about that mindset: how do I enable the team, how do I clear the roadblocks for them, when somebody really needs help, how do I help them? That mindset is what helped me. Did I have it from day one? Absolutely not. Sometimes it's difficult learning; sometimes you learn through experience.

Victoria Mensch: I'm really interested in your view on this, especially around empathy and the servant leadership approach.

Harish Epuri: As a leader you've got to make some tough decisions, especially with productivity goals and challenges. My thing is: don't commit very aggressively without talking to the team members. Or, if you commit something aggressively because you had to, because it's an important thing, then be very transparent with the team about what circumstances made you commit. It's not, I told them I'll get it done tomorrow, let's do it. You're not going to win the trust or the respect of your team that way. But if you have that relationship and trust with your team, then even if you commit, they will rally behind you, because they know you are the person who cares for them. You don't need to ask them to work extra hours or weekends; they will rally. The team should feel the leader is backing them always, no matter what. That's what makes the leader's job easy. As I said, you earn trust by drops but lose it by buckets. It takes a long time to earn that trust, but you could lose it in one shot with one misstep. And it happens both ways, not one way. Each person is different, so as a leader you have to adapt to each person's style.

Victoria Mensch: What benefits have you seen from this approach in team performance, retention, trust and execution quality?

Harish Epuri: If you have trust, your focus moves away from guessing things, and your mindset is only focused on execution, because you are not worried about whether this person will leave. The trust is there, and if something is wrong, they know how to call you and you know how to reach out to them. There is no ambiguity. You don't need to set a guideline like, you can't call me after 5 p.m. That whole communication channel is established, so you focus on execution. To me that is very important. When I come on board at a new company, one of the things I try hard at is understanding people and working within their style. It's also very important to be very clear about what you expect, because I can think something, but if I don't express it, it doesn't come out right. In the heat of the moment, a lot of times we forget to do these simple things. So it's very important to have that communication protocol established. Once you establish the protocol and stick to those principles, the natural byproduct is trust.

Victoria Mensch: If you could leave leaders with one principle for making better decisions through transformation, what would that be?

Harish Epuri: I would say two things: listen with open ears, and give the benefit of the doubt.

Victoria Mensch: Fantastic. Thank you so much for sharing all your years of engineering and leadership experience with our listeners.

Harish Epuri: Thank you, Victoria, thanks for having me. It was great having this conversation. I really appreciate the opportunity.