How AI leaders are preparing for the future

My name is Junie Dinda. I run product marketing here at Asana, and I'm joined with a very esteemed panel who you might have seen during the keynote, but I want them to introduce themselves quickly again.

I'm Max Poritsky. I'm the chief revenue officer at Zymogen.

Laxman Nathan. You go by Lax. I'm the CIO at Skydance.

And I'm Michael Warden. I'm the associate vice president for global service delivery at NYU.

Excellent. So I think what we saw at the keynote was, like, a little bit of a preview into how AI has been transforming their their organizations. So we're gonna take a deeper dive into this and just, like, ask some questions to it kind of family style, popcorn style as well. Okay.

First one's a three parter, so let's, let's buckle up here. Okay. So so so where is your organization today in their AI adoption journey? What workflows are you tackling first?

And how has that started to change the way that your people work from day to day? I'll start with you, Max.

Sure.

So we're a dietary supplement in this... Company, and so we're not a tech forward company sort of as it is.

Our journey, we're still very early in it. And in fact, I sort of have had to run this side kind of experimentally like a skunk works almost.

But, you know, it's... We're finding that people are sort of organically adopting AI in different parts of their day to day workflows. We're still working on sort of a a top down strategy. We just don't have it in place yet. So instead, we've sort of run this, like I said, like a skunk works and just sort of project based, improve concept, and show that we can drive value to to our customers.

That's right. Skunk Works projects do move fast as well. Right?

They move fast, and that's that's how I like it.

Very nice. Very nice. Cool. How about yourself, Luch?

At Skydance, we've gone through...

I'll I'll say we've gone through a lot of change over the last couple of years. We just changed our name recently and became a joint company here with Warner Brothers Discovery. For us on the legacy Paramount side, we have been using all the AI tools you could think of over the last two years. But I think over the last six months, what we're seeing is more of a change in culture because of AgenTek workflows, but even just the tools of building, like using Cursor, Claude.

So over the last six months, you know, we're seeing, at least on a product tech perspective, like new energy about, like, how fast we can build these tools. And on the other hand of it, like, a lot of our users who have been using the tools are now seeing, hey. Look. We can do projects much faster than we've ever had before.

So that's been a real change in how we look at how we do our work.

Excellent. Excellent.

So NYU is a kind of a unique, story in this aspect because at NYU, there's such diversity that if you you can think of almost any AI use case, somebody's probably doing something with it. We have faculty who are researching and advancing the concepts of AI itself. We've got students that are hungry for any tools that they can use appropriately To to understand what AI can can help them do. But there's a huge appetite.

I just saw a student project recently, for example, where they were trying to solve some of their own problems of, I really wanna know if there's any room in the library, so let me vibe code something that I can pitch to IT to try and talk about what...

How we could communicate available space better across campus. So that appetite is great, except trying to do that safely is the challenge we're really navigating right now. There's a huge appetite, but there's not great guardrails. And we're such a diverse global organization that it's hard to to stay aligned across our Abu Dhabi campus, our Shanghai campus, what's happening here in New York.

And so right now, we're still just trying to get our arms around what is it that we enable, what do we steer towards, and how do we keep everybody safe while we do it.

So kind of putting kind of like the foundations in place, getting the the early, the learnings and everything, and then moving forward from that.

Yeah. We're trying to create safer spaces for people to pursue pursue that AI. So when we think about what we might do with Asana, we wanna make sure the platform is stable and secure, and then intentionally pursue what we do in that space rather than having something go sideways on us external to that.

Interesting. And then I think, like, like, something's interesting just about the overall, like, foundation that you guys are thinking of building. How do you, like, you know... How have you approached or how will you be approaching it, like, building that foundation at the organization? Like, at what teams, like, you know, does it start at? And what's your kind of approach? I'm gonna go in the middle.

I'm gonna go up So I'll start with, like, one is actually the opening up the enterprise applications themselves.

So similar to Asana and some of our other applications, having guardrails and ability, but being able to connect into those tools and make sure that the we have access to it. But the key thing for us on the foundation is that it's the data itself. So you can have all the agents in the world that you want, but if your data is not good, they fail.

And so in our strategy, we're kinda looking at that middle layer of saying, how do we build confidence in our data Okay.

Even in when we're connecting multiple systems together across the board? Because, again, we're similar to NYU. Like, we have systems all over the place. We have a lot of enterprise, but then we also have some... I'm gonna call it more business specific applications. But, like, our our kind of investment right now is to really sit down and say, how do we use AI to actually normalize the data?

And then if the data is solid, then we think the agent process will end up being That makes a lot of sense.

It's just like, you know, if if the data is not clean, the workflows won't be clean.

The processes don't Exactly. Don't get cleaned and they stop.

So, Max, I saw you nodding vigorously there.

Yeah. I mean, data is really important. We're a much smaller company.

We've leaned into subject matter expertise, really understanding our customer and our... And the process and sort of where the friction points are. But even as we've deployed agents, we've then run analytics on the interactions with those agents in order to surface common themes, questions that come up, and friction points even within the agents. So it's allowed us to iterate in a really sort of clever and quick way, but that's that's very much based on the human experience.

That's amazing. That's great to to do the analytics and be so data driven as you're trying to go through the process and, like, you know, optimize. Amazing. Michael.

Yeah. I think what... NYU really started, I think, our first big AI effort was actually in service to our researchers because there was a big appetite for the researchers to get their hands on and start advancing some things. So a couple years ago, we really started standing up an environment in which we could steer that activity into some of the tools that we're enabling to support AI. But since then, the administrative space has really taken off. So almost every vendor is talking to us about what their AI answer is, and we're trying to make good choices about what we pursue there. But we're... I think we're trying to stay in safer spaces. So we're not trying to start with our most sensitive data or most high risk data.

But it's interesting. Almost every team is doing something to to lean into what can make what they do better. We're having the exact same data challenges. We think some of the AI capabilities that... Some of the AI capabilities will actually advance our ability to do the data governance we wish we would have done for a long time. Because all of a sudden, it becomes very real how messy data shows up in a workflow. And the appetite to to tap into those AI workflows is so high that they're saying, do we need to do to get those things in line? And it's it's helping to advance some of our our data governance and our enterprise information management approaches.

K. So I'm hearing data. Data is a a theme. Analytics is definitely a theme there. So I'm gonna go into a little bit of maybe key learnings or, a war story. So is there, like, you know... Like, getting the foundations is the right first step.

Is there something that that you recognized in, like, know, hindsight or something that broke that you might consider, like, know, doing... Like like, any key learnings from that?

I'll take that one first. Okay.

One of our... We had a a challenge we're actually working on with Asana right now is to try and automate what we thought would be a really straightforward thing to automate, which is how do we do our AV events. As I'm looking around at all the the the technology here, it's impressive to see everything that goes into that. But coordinating where our staff are available, what equipment is available on what days, coordinating billing, recharge, all those processes.

We jumped into, can we create a... An integration or workflow that does that? And we made a mistake that I think a lot of tech companies make, which is we forgot about how important that that personal or tribal knowledge is for the staff involved. And so we kinda had to slow down and make sure we were bringing our staff along to be comfortable with the idea of automating that process, which is true with any IT. But with AI, there's a... There's some fear to it too. So we're trying to be really careful and not go too fast, not chase the the new technology without staying true to the fundamentals of putting the people first. So that that was a big lesson for me.

I'm gonna say we we made zero mistakes and everything's been...

No. No. Great. You know.

No. I mean, you know what? What's actually a different angle for us is that because of, like, all the AI, like, coding tools, I'm gonna go back to that for a second, is that people look at that and they're like, oh, wait.

I can build things so much faster

That I would say that, like, it's not... It's it's more just a lesson learned. It's like when you start building this stuff, everyone forgets the whole basic principles of product requirements and all those things. You're doing these faster iterations.

And I think what what we're seeing though is that we have to sort of adjust the old way we used to, you know, deploy code

Versus how we're doing it now. And actually getting our business into that, like, more of that product mindset is actually the lesson learned, is that we kinda have to train them to almost be like product owners because they, you know, they could see fast iterations. Your first iteration might look great, but then it's your second and third where you're actually trying to make it reality. Getting them to start thinking that way and actually even on the product tech side, from our perspective, is adapting to that. So there's a... It's just a cultural change. It's like that... And I'm still saying that we're still learning in how to develop it, but because of the it's a new thing for everybody.

Yeah. I think that's a great, that's a great point. I think building building something quickly is not the final finished product. It might actually just be a prototype, but then you might think, wait.

It looks great. It it it works in one or two use cases, but then can that scale? So Yeah. I think that's a great mindset.

Yeah. I mean, you know, I think, again, coming at it as a nontechnologist, I wasn't sort of bound by the understanding of the limitations.

So it was more like I'm dreaming it, and let's push it to the point where, you know, those limit... We hit those limitations.

That don't necessarily break it, but let's push it to and beyond its limitations and iterate from there. And, honestly, like, the stack team has been phenomenal because we've done that multiple times where they've come back and gone, yeah. Give us till Thursday and see if we can make that work. And I think that's been part of the the partnership we've enjoyed with them.

Oh, that's great. That's, like, the good responsiveness and the building on both sides to build that partnership.

Amazing. Okay. I'm gonna shift gears just a little bit. Okay. As, like, know, more work shifts to AI, how are you actually governing who is accountable, and how do you keep the visibility into what is actually happening across your organization, across the projects?

So I'll...

Can take that one.

Yes. So, I mean, I think when AI first started, and given, like, in a media company, it's... We have to look at it from a couple different angles. We have to look at it from our content side. We have to look at it from our enterprise side. And so early on, we came up with...

Typically, think a lot of companies have done this, but the the AI governance with your info sec, privacy, legal teams, in our case, like employee relations, HR, finance. Having... We started pretty good with a... Having like a a nimble council That was adapting as we were doing that. And so I think what what has probably changed more is because the technology keeps changing faster and faster, having just more visibility into the tools and What's being in the environment.

So I think the fact that we have all the right people having an active dialogue in this

Has been really good about the governance. I think going forward, I think you're gonna see more tools that's gonna be used for monitoring this in your enterprise stack, and I think that's gonna be where where the road's gonna go with this.

Yeah. Exactly. As you would like increase the amount of tools that you're taking and everything from that perspective. But Yeah. I think it makes sense to have that that cross functional team that brings in different perspectives, especially in this, like, this, like, you know, newer space that's constantly evolving.

Yeah.

So that makes sense.

Michael.

Yeah. So we've done the same thing. I think most organizations are trying to pull together the right slice of leadership to to weigh in on these things. We're really starting also by just trying to establish some really clear guardrails on what not to do.

It's it's a little easier to to think about these are the areas where we we can't, you know, violate existing policies or violate existing capabilities. Because there are things that AI can do. One of my favorite examples is, you'll see a lot of products advertising the ability to, you know, to listen into the room and take notes for you. Right?

You have to be really careful here in the state of...

Or in the state and the city of New York where there's there's some really specific legislation around around some of those things. But people will will jump to using some of those things. So we're trying to educate and provide those general guardrails. Where we're trying to control from a technology standpoint though is is, again, it comes a little bit back to the data. Doing things with with AI tools on their own is one thing, but controlling where and how they're integrated with enterprise systems is really kind of a control point we're focusing on. That trigger of will I grant MCP or or or access through this particular integration to the most sensitive data is really where we have to be as a as a really distributed university.

And then as...

Is it about then communicating that and and ensuring that that those guardrails are actually, visible

And accessible to people so they can work it.

We're trying to tie it into things like our procurement processes, for example. That's often a a good control point for us to make sure we're doing a good assessment of what AI capabilities are coming in and the things that people are trying to purchase. But even that, we're also just trying to detect what's out there. So many products are adding AI elements that are that are implemented in a product that's already deployed. So it's also just keeping an eye on where each of the vendors or products jumping up in some of these spaces and trying to trying to detect detect where we can, what we can.

To the accountability question though, that's actually one of the things we focus on is if you're a system owner or a business owner, do you understand that accountability? And if you don't, don't be in such a hurry to try and deploy something.

And then that's Sage advice. Definitely. Yeah. Thumbs up there. Max?

I think the only thing I would add to that, you know, just a general sort of, reactive governance that that our tech teams have taken, is that the the specific AI agents and tools that we've built, particularly with the stack tools, are very user specific. So they're built for a particular group of of users within the company. It's either the sales team or it's a a customer group or or or FP and A team. And so, you know, that in itself sort of gates who's using it and how they're using it.

And, we then go back on a fairly regular basis and take a look at, you know, the the use cases, how people are interacting with it, what it's pushing back, and sort of gauge, you know, the quality of of those interactions and and learn how we can iterate from there. And if there are areas where we need to, you know, add additional governance or even limitations in in the data capable. Because right now we are...

We're integrating multiple systems and it's all very new. We're pulling from our ERP and we're pulling from, you know, our CRM and we're pulling from, you know, just documentation that we have on on product and and IP and such. So it is very much a learning in progress as we go.

And then it depends on then, like, the workflow then, the team, the use of the data, the type of data Exactly.

That's actually being defined. Okay. That makes sense. Okay. I'm going to a metrics question that everyone wants to know on.

How do you know if it's actually working? How do you know that if you've been successful at, you know, implementing AI and, like, you know, realizing that promise of AI and, like, the productivity? Gonna go to Laksh because you're smiling.

It depends on who you ask at our company, to be honest.

So so it's easy for, in some cases, product tech because you could say, I'm measuring based off the usage and tokens of the tools. But the the harder one is on the business side. Because, I mean, typically people are gonna talk about ROI and, like, productivity.

But what I've actually found is more qualitative in some cases because similar to what you mentioned, like you're working with specific department heads and business units. And a lot of times what we've seen is in the...

Especially in the last year is that they've got like three goals and they really need to like make their life easier. Right? And I actually think that like getting the department head actually to be excited by the outcome of it is like, it's it's hard to even describe that because you can't always put a metric in front of someone and say, hey, you got x y and z percentage. So a lot of my times is like, if I'm gonna be spending some time and like working with some department heads, I want I want it to be a two way street.

I want them to really be like, this is going to change the way we work. Sometimes it actually...

It's more like we need to implement this and then we need to figure out the metrics afterwards. Because then we actually can see how we're productive. Because we have a theory that we can do it, but now we have to actually use the system to make it happen. And because it takes faster to build these things, it's a little bit more easier to to convince people to just try it Yeah.

Before you have to convince them of an ROI like you used to do it back in the day.

Yeah. Just to dovetail onto that, I think that you bring up a great point, which is it it is very much subjective. Like, the feedback is in adoption. You know, that's the metric I look at first is across my sales sales organization, how many people are using and how much are they using this tool?

And how much am I hearing about it from them? And that's that's a big part because that's... That perception of value is the most important component of gaining adoption.

And then the metrics, think, will come later. I mean, we are sort of a little bit...

It's a leap of faith.

Yeah.

Yeah. Know, that it's having a meaningful impact. But when we have a customer facing tool, that, you know, we we know saves them a lot of time Time is that... That's that finite, you know, resource that we all have, And and and that's very measurable. And so for those kinds of tools, time savings is as good as anything.

Okay. Interesting.

I'm gonna totally pick up on the time thing. So to me, I I don't think we can measure. I like I like not to talk about ROI, for example. Chief revenue officer wants to talk about ROI.

But to me, it's it's are we are we accelerating towards the outcomes that we're accountable for? So I'd like all of our services across IT to be thinking about what are the are the needs we're trying to serve and what are the outcomes that we're trying to create. And so ideally, wanna be measuring, are we getting to those outcomes faster or are we creating more momentum or more agility in the organization? But we won't be able to measure that in in...

For a while. And so a lot more is, do we feel like we're moving in the right direction? Are we reducing some effort? Are we...

Is there an appetite and we're serving it? And to me, it it comes back down to, are you able to do more with with the tools that you have? Are you freeing yourself up to to spend time doing the things you would rather be doing? Like, one of the things we'll often talk about is, I think AI can make easy the work that everyone procrastinates and doesn't do.

So creating knowledge articles or or updating the entries in our in our catalogs or updating websites used to take a lot of time. Even, you know, writing a really good project charter would would take a while. Now, we can create the first draft really, really quickly. So I'm hoping it's it's it's in producing higher outcomes and higher momentum for the teams.

Like Cicero said that time is a resource of inestimable value. And so how do you how do you value time? You just create more of it. So...

Yeah. I think the perception of value, I think that comes in, like, you know, it it it comes into play when you start to think about the adoption and and and one team or one person's getting value, and then you get other people more more curious about it. Like, you know, like like, have you seen that in your organizations where, like, that one team started to, like, you know, drive the momentum to start experimenting and actually adopting AI?

Yeah. But I... I'd say we sort of cheated in in that, you know, I only built something for our sales group initially. And so they were the test bed. And and from there, once once we saw that it was infectious, they were finding, you know, a lot of value in it, then I immediately had other areas of the business I could sort of point it towards, and and you just build that momentum.

Yeah. I think IT tends to be... You know, if we're trying out something, you know, within the IT organization, sometimes that's that's a place of generous momentum. I think AI has been a little different in the sense that folks are are finding tools on their own and they're coming into work saying, have you tried this thing? Have you seen this thing? I started seeing it when all of a sudden all the slides started to not follow NYU's standard look and be.

And all of a sudden, they still have this great purple branding and these these nice things.

But I was like, alright. People are starting to create some things. And when you when you see one person do it, somebody else is getting curious. And there was a lot of just really organic adoption. Again, I think our challenge is getting out in front of that and meeting the need before folks go solution on their own on something risky.

But it really popped up in lots and lots of different places. The teams that I think have made the most momentum though though are the ones that have a lot of repeatable work and feel overwhelmed. They're the ones who seem most interested in grabbing and and and providing relief.

I agree with you definitely on that, particularly in that last statement about the folks who have the most number of work and like they they... And they've been dabbling with themselves. So I I always find that like the successful ones have been... They already have self trained or they've been using the tools. They understand the concepts.

A lot of them might build something a little bit on their own. And it actually makes a nice bridge with us because at least they've done their...

I almost wanna say their their business requirements, you know Yeah.

Like themselves.

And so it's a lot easier for us to have, like, an engineer or some configuration person to sit with them and say, oh, yeah. We can turn this from your computer to an application. And depending on, you know, again, with same guardrails, things like that, that's, like, some of the challenges with the enthusiasm of some of this is like they're not used to that enterprise kind of thinking. So they might build it just for themselves.

And so so to be fair, it's like I think that's also opened up doors for us because someone comes up with a really unique idea

We kind of take that and kind of turn it into an enterprise solution and then line it up with the outcomes associated to it. So it's a it's a give or take that we've been seeing.

Yeah. That's like the like the typical journey of like, you know, consumer consumer, like, you know, software applications kind of back in the day of, like, then using it and then wanting to use them in the enterprise. But there is, like, that distinction between, like, you know, different governance, different everything like that. But I do think that the adoption of consumer, like, you know, consumer agents and AI tools is is making people more curious.

Yep.

And and even getting to that thing as as, like, you know, Dan was speaking about, hey. This changed my life at home.

I want this to do the same thing at work Yeah. And trying to find that that solution. Amazing. Alright. Let's talk about AI being rolled out broadly. So, we've kind of, like, touched on it in, like, you know, different parts of the organization building that foundation.

Like like... So, like, what is that point in where you're confident that it's, like, you know, ready to go, like, you know, broadly across the organization? And and then on the flip side, like, you know, at what kind of, like junctures do you think, hey, wait. I think we need to stop and test or or just wait a bit.

I think it's a...

One, I would say is the type of data you're rolling out.

So depending on the business process or the group that you're working with, making sure that like, what's the precision of success? So in some some groups... So two parts, data and the precision of success.

Okay.

So the data is literally like, hey, if you're gonna get like a... You know, you're not dealing with sensitive information, you're not dealing with anything, that's a good gate because that kind of probably indicates you can move a little bit faster. If it's... If there is more sensitivity to it, just having that rigor of making sure is everything protected, everything... Is the output exactly what you're expecting it That's to one science. But then, just generally the...

I'm gonna call it more just the...

Just that output.

If if you're dealing with the precision part

Finding that if groups are accepting less 100... Than 100%.

So imagine when you deploy something and they're willing to iterate the outcomes of it. That's when I feel like there's... That's that's one avenue.

Okay.

So if you get to... You you have, like, the human in the loop. You're looking at it. You're always consolent and improving it over a period of time. That's a successful piece.

But then if there's a precision where you need to be a 100%, that's when I I get a little bit more conservative about like Got it.

Are you really gonna roll this out? Are you really gonna... You know, so I think it's a different mindset with rolling out these tools because you're so used to the old days of like, it's either right or it's wrong.

And now that you could see that you probably can learn Yeah.

From doing it, that's that's a new mindset that... But it also depends on the, you know, what's your risk factor and your level and that...

It's kinda like a growth mindset From that perspective. Exactly. Makes sense.

Yeah. Think it's a great point. It's it's it's no longer binary. Like, it works or it doesn't work.

Yeah.

You know, for us, it's very much about... I think you you pointed it out too. Like, the quality of the data that it's bringing out has to be there first to feel confident enough. And then I'm a big fan of the test and learn of the pilot. Yeah. So I'll find a couple of, you know, key people who I think, you know, really serve either a legacy employee who really has that domain knowledge and let them play with it for a while.

And then we'll expand the group out before we go sort of full full deployment.

That...

That's a big part of ours. And the the kind of pilot and then deploy model is is really what we're doing. But I I would just add as a wrinkle to it, we kinda need to publicize those pilots really broadly so people don't go experiment on their own. They're like, oh, somebody's looking into it.

Because, I mean, we had so many different folks, you know, reaching out and saying, hey. Do you have a phone number for somebody at Claude that we can talk to because we wanna buy it? We're like, tell us more about that. So so doing those pilots is a way to socialize that we are doing something, but also giving us the time to assess it before we are really ready to roll things out more broadly.

So we're a big pilot then deploy group.

Yeah. And making that visible makes sense. So, like, you know, they know that that you're thinking about it, that you're invested in it, like, you know, as well. So they have, like, the, like, kind of the peace of mind that someone is taking care of this so they don't have to go rogue.

Okay. Makes a lot of sense. We're almost at about time, so I wanna leave you just with, like, any any advice to to everyone out here on, like, the the top learning that that you learned from, like, at this point in the journey of, like, you know, rolling out, like, know, AI transformations? I heard themes of data.

Heard, like, themes of iterating and growth. Anything else that you wanna leave the audience with?

I'd say, I think you have to create safe spaces for people to do these things because in the absence of... I I like to sometimes use the analogy of a of a playground is that you wanna create a place where there aren't gonna be really sharp edges, and it comes back to the data or the governance. But if you don't create those safe places, people are gonna go try things on their own. So I think one of my biggest recommendations is we won't be able to sate the appetite to chase and try out these new things, but we have to keep people safe while they do it. And so, for example, there's a lot of people vibe coding. And we're interested in in creating a kind of a canned, almost automatically deployable space for them to put that. Right? So the folks who are now doing that coding don't know about how to use GitHub and how to use all the different tools we want to do good enterprise release management.

So I feel like we have to advertise the the safe space to go do it more proactively. That seems like a big gap for us.

Yeah. Safe space to experiment. Yeah. Great.

Like your agentic registration kind of process. That's definitely on the same boat as us. But one big thing just from an AI transformation, which I didn't really mention, is just actually getting the top down buy in. Yeah.

Because I really...

I know it sounds really simple, but, like, having your leadership really buy into, like, some sort of reimagining of the process. It doesn't have to even be AI. But knowing that AI is a is a tool that can help facilitate it. It it gives a good top down approach for everybody because I think that's where...

If you know you're in a protected space and you've got the ability to make changes

It it really does trickle to all of our departments across the board. So getting that top top down leadership is probably huge from from my perspective.

Definitely. Thank you.

I guess what I would say is don't deploy AI for the sake of deploying AI. Yeah. Yeah. So lean into your domain expertise.

Bring the people who understand your customer, your process, your industry the most. Bring them to the table and ask them, what is the one thing, the one friction point that if we could alleviate that would make a big impact on your ability to deliver value to your customer? The one thing, simplify it and then solve that problem. I think that's that's the approach from, again, from a nontech guy.

I think that's what I would probably...

That would be the advice I would have given myself a a year ago, probably.

I think that makes sense. I mean, at the end of the day, like, it's it's not just about using AI. It's it's about trying to solve the problems. So it's trying to figure out... It's mapping out the workflow, figuring out the jobs to be done, and then figuring out how to get to the outcome. And I think that's I think that's a great point. Well, thank you all. This was a great, great discussion, And and thank you for the participation in the keynote as well.

Thanks for having us. Yeah. Thank you.

Thank you.