Leveraging Agentic AI: Key Strategies with Workday’s Shan Moorthy

Paul Spain:
Hey folks, greetings and welcome along to the New Zealand Tech Podcast. I’m your host, Paul Spain, and privileged to have Shan Moorthy, Chief Technology Officer for APAC at Workday joining us again. How are you, Shan?

Shan Moorthy:
Good, thanks. Good to be back. Actually, almost a year since I was here last. Lots of things have changed.

Paul Spain:
Yeah, it’s been very fast moving on the AI front, which was a big part of what we spoke about last year. So I’m really looking forward to delving into that. Maybe you can just remind listeners where you fit in and what your life in the tech world looks like.

Shan Moorthy:
Yes, I’m the CTO for Asia Pacific for Workday, and one of the coolest things is that we’re on the bleeding edge of this AI revolution. In particular, one of the things that excites me about the role is that I look after the engineering that’s really redefining how we work. And considering that most people spent a vast majority of their time at work, it’s kind of defining what their life looks like, which is fascinating, like what’s changing and what people expect it to look like, what people are hypothesizing it’s going to be like, and what reality looks like as well.

Paul Spain:
Yeah, gotcha. Of course, big thank you to our show partners, Spark, One New Zealand, 2degrees Workday, Fortinet and Gorilla Technology. Really appreciate the support that keeps us going through the year. Just in terms of. There’ll be some listeners from a context perspective who probably benefit from just an overview of Workday and what you do and where you fit in the market.

Shan Moorthy:
Workday is we manage. We help organisations manage their people, finances and agents. Now, traditionally we were people and money. So in an organisation, those are two biggest assets, your people and your money. It’s how you get work done, right? So you have the capital, the resources to get work done can either be labour or it can be the finance that sponsors it. What’s happened of late is with the advent of agentic AI, we work on this notion that the future of work will be done by either people or AI agents. The work itself doesn’t change. The work that an organisation needs to do in order to service its constituents or service its customers or you know, its clients or students even that doesn’t change.

Shan Moorthy:
The sum total amount doesn’t change. But who does it? How do they do it? When do they do it? That is up for grabs at the moment. And one thing that I dropped this into any conversation with an executive that I have is that they are the last generation of managers that are going to be managing a purely human workforce. Every generation that comes after them, and probably them included, is going to be managing a hybrid people and agent workforce in order to get work done. So Workday really is a platform that allows that hybrid human agent ecosystem to thrive and it provides the rails on which you can do that safely as well.

Paul Spain:
Yeah. Really keen to dive in and I guess pick up last year’s conversation and get your thoughts on how things have progressed. What’s working. Yeah. What’s not, what are the things we see Maybe disappointment with AI and just some of your insights because you get to interact with a lot of organisations across the region and beyond. Maybe we can start just with your view on how you’ve seen AI progress over this last year. What we’re three and a half years in since, since sort of the, I guess the chat, you know, what started with the chat chat GPT era that I, you know, I think really started giving the tech and business sector, you know, a bit of a wake up call around AI. Of course, AI been around for, you know, a really long time, but that’s kind of the moment we started.

Shan Moorthy:
Yeah. I think realizing, I think last time I was here, I was joking that in the 90s I did AI as a course.

Paul Spain:
Yes.

Shan Moorthy:
And, and I got told my job prospects was, was recognizing number plates on the freeway and that was it. And it’s come a long way since then. Look, in the last year, I think this time last year we were sitting here talking about Gen AI. Right. Like that was the panacea for everyone. Yeah, yeah.

Paul Spain:
And agents.

Shan Moorthy:
And agents were just picking up, but not in the way that we’ve seen it develop. I wish I had a crystal ball at that time and could predict the hunger that a lot of organisations are now expressing for having agentic behaviour. And there’s a few milestone moments, I think along the way that where, where, you know, culturally we, we sort of became okay with autonomy to some of these agents. Definitely last year we were, you know, don’t give it to the bots. Don’t give it to the bots. And, and we were even talking about is it really an agent or is it just an RPA that’s dressed up as an agent because it has a, a chat LLM stuck on top of it?

Paul Spain:
Right, right. You’re talking robotic process automation and those tools which of course have been around,

Shan Moorthy:
they’ve been a long time. Yeah, you know, about a year ago. That’s, that’s kind of. We, we had a lot of Gen AI and we had robotic processors which had an LLM based chat interface in the front of that rpa and that was, that was where we were at. Rather than talking to the technology. I can tell you what CIOs and CTOs around the world experience, which is they had a rude awakening when their CEO came up to them and said, hey, I’ve used ChatGPT at home, my kid uses it. What’s our AI strategy? Tell me. And they furiously research this thing about GenAI.

Shan Moorthy:
There’s some, there’s some productivity claims made that, you know, me included, that, you know, certain productivity numbers will be hit. But Gen AI will only take you so far because it’s generating content based on what it’s seen before. Yeah, which is not a bad thing when it comes to the right use of it. Like if you want to create content, if you want to generate something that you generate thousands of times over and over again, then yes, Gen AI is the perfect tool for it. But I think a lot of organisations actually signed up for a huge productivity uplift on the back of Gen AI and it didn’t really come through. I think statistically a lot of research was done And I think 5% of those proof of concepts actually made it to production, which is millions of dollars spent in experiments that never actually made it to production. Unfortunately, that meant that we went into this trough of disillusionment where, you know, everyone became a bit sceptical about the, the productivity that could be unlocked with AI because there was no, or not much awareness of the difference between Gen AI and agentic AI at that time.

Paul Spain:
Yeah, yeah, just you mention, you know, proof of concepts, POCs. This does, you know, does seem to be with lots of technology, especially new technology or organisations, you know, interested in adopting something new. Yeah. An important part of the, you know, the journey is that, you know, trial and experimentation again coming up with some way of, you know, proving things out. Are you, have you seen any particular patterns in the types of POC projects that organisations have been doing or has it been massive variety?

Shan Moorthy:
Look, it varies based on industry and varies based on organisational size as well. I think the recipe for a successful proof of Concept is where the outcome is known or is well defined. Where it goes wrong is when there is a smorgasbord of tools made available and then a solution is found for. Well, a problem is found for the solution rather than the other way around. Where an organisation is very precise and can articulate this is the end goal that we need to get to and, and then give creative freedom to the engineering teams and the IT teams and even the business units that are involved to then adopt any kind of AI that is required to get to that end state. That’s where we see those proof of concepts then move into production. But what happened a couple of years ago was pretty much every single organisation came out and said we have AI. If you didn’t have AI in your brochures, you weren’t selling anything.

Shan Moorthy:
So there was a huge amount of tools available and it wasn’t too far fetched to have more than one tool in your portfolio. So then there was confusion about what tool do I use, where do I use it, how do I use it? A problem I’m trying to solve, but that’s kind of, it’s tapered off now. Where we’ve now coalesced around defining outcomes and having AI that supports those outcomes and working in that way.

Paul Spain:
In terms of the tools that you’re and mechanisms, there’s obviously lots of ways of doing AI referred to. What are you tending to see out there amongst your clients? Which I’m picking would be leaning towards larger, larger organisations generally. But you also work at the, at the smaller end of town. But I imagine you probably have more conversations with sort of, you know, bigger businesses and you know, government types of organisations.

Shan Moorthy:
Yeah, look, we have everything from governments to health departments to education, Auckland University, et cetera. There’s a wide variety of organisations. The one thing is that we are a very engineering first principles company, which I love. One of the first principles is we only maintain one version of software. So there’s no forks of the software, there’s no version for the bigger customers and one version for the smaller customers. We run it all on the cloud and every week we deploy whatever is at the, you know, the main branch to production. Rain. Hello, Shine.

Paul Spain:
Yes.

Shan Moorthy:
If you miss the train, just, you’ll be there next week. Right, so. But what that means is that the software that we develop is built for the bigger end of town, but the benefits trickle down to the smaller companies as well. And what we’re seeing, just to sidetrack a little bit, what we’re seeing is the rise of smaller Companies that are punching way above their capabilities by using the agentic capabilities that are available to them as well. And I think that’s an expectation that a lot of organisations have, is that it’s not just about cost out anymore. It is, can we scale up to the next level? Can we service people better? Can we take on more ancillary tasks that we might have shied away from because we didn’t have the capacity to, or we didn’t have the creative space to explore?

Paul Spain:
Yes, yes.

Shan Moorthy:
So that’s a roundabout way of saying, you know, the, the organisations that we’re, we’re talking to are, are looking to grow. And there’s an expectation from the executive teams of most organisations that their people are able to operate at a speed and scale that we’ve never seen before. So it’s not just about productivity anymore. It’s actually about unlocking a new category of organisation.

Paul Spain:
Yeah, that’s exciting. Now you talked about not just having the tools and then looking and working out, oh, what can we do with it? But looking at the business problems and the opportunities. How are you finding organisations get, get the most success in terms of, you know, defining and finding those opportunities to, to leverage AI? Because you can, these things can be pushed from all sorts of, you know, directions. Right. You’ll get the, you know, a board or chief executives, like we’ve got to be all in on AI, but without necessarily understanding, you know, where it fits. And then there’ll be other people in different places who will see different opportunities. Is there a particular methodology or approach that’s worked really well, that’s stood out for you?

Shan Moorthy:
There’s a couple of things there. So one is with the advent of what I call general AI, which is your chatgpts and your Anthropics and the Clauds and things like that, and Gemini, for example, there’s a notion that you can build your own agents and you can solve everything in your organisation. And certainly the temptation is there.

Paul Spain:
Right.

Shan Moorthy:
But what you end up doing is you build a shadow ERP in the process. Basically you take what an ERP is supposed to do and you just build small bots. Now you’re maintaining an erp, which may not be like what the organisation is there for.

Paul Spain:
Right. So, yeah, lots of people potentially building out all, all sorts of things that are doing what you.

Shan Moorthy:
Yeah, and this is sort of symptomatic of having access to tools without a clear vision on where the tools should be applied and where the business needs to evolve. So the more successful organisations identify where they make their money or where they are supposed to succeed. And you can kind of divide that into domains, right? There’s a domain of managing people, domain of managing finances, domain of managing customer relations. These are all in technology. We have domains for these things. If you decide to build all of that using general AI, you will effectively rebuild software from scratch. And then you are taking on the onus of maintaining that software, maintaining and supporting, which may not be the goal of the organisation. So if you think of it as an 8020 rule, 80% of that is commodity things that can be solved using a platform and what I call vertically, right? So this is AI agents, for example, that are built, which have those common sense rules, have the guardrails built in and so on.

Shan Moorthy:
So in all of those domains, the best thing to do is to identify a platform that gives that out of the box, leverage that and then identify the 20% where you actually differentiate your business from other businesses.

Paul Spain:
Yes, yes.

Shan Moorthy:
And then build that part only rather than trying to build a whole stack from scratch. So at Workday we talk about it as probabilistic AI and deterministic platform because most organisations we work with govern, you know, have a governance system, have, are regulated. They have policies or laws that apply to them, have a watchdog maybe or an auditor that will come in and inspect and ask for receipts, right? So do you want to use general AI to build all of those things out or would you like a platform to give give you most of that out of the box and then have the IP that your business represents be the one that you innovate on. And where organisations are succeeding is where they actually leverage a lot of that and allowing them to maintain that deterministic rails which is required to operate because take for example payroll, right? It’s not okay to be 98% accurate on your payroll. Not at all.

Paul Spain:
Right.

Shan Moorthy:
Because there’s 2% that’s going to get their checks wrong and that’s life changing for people.

Paul Spain:
So that’s where you need the deterministic technology that is going to be consistent the same every single time.

Shan Moorthy:
It’s rails. And really what we’re talking about is when we talk about speed and scale, think of it as the metaphor that I always like to use is a train. Think of it as going from, I don’t know, steam or diesel to electric. Right? We’re going from a way of working that’s able to deliver that output of work much faster, but you still need the same guardrails, regardless of whether it’s a person doing that. Work or whether it’s an AI agent doing that work, you need to know, are they qualified to do that job? What skills do they have? What makes them eligible to do that job? Who is doing the job? What was the performance like? What was the outcome? Under what conditions did they do the job? And do you have an audit trail that can actually prove all of these things? So those are non negotiables and those are things that you don’t want to be building as an organisation. You want to be focusing on the, what is the job that you’re trying to do rather than all of these other things, the scaffolding around it.

Paul Spain:
Gotcha. Yeah. How would you sort of split the things that fall into being trustworthy to use probabilistic AI on? Because obviously there are certain things you need a very clear and consistent result on. There are other things where, you know, I guess the, where the AI, you know, can come up with maybe somewhat a variety of, of results and sometimes not exactly 100% the same every single time.

Shan Moorthy:
So one piece of advice that I always give is that it’s a couple of statements, actually. So the first one is AI can never generate the truth. It can give you an opinion based on things that it’s seen, but it can never generate the truth. Only people can generate truths. Right. And that’s why we need the human in the loop. And actually, when we first started on this a couple of years ago, we were talking about human in the loop. I think we’re finding out that the humans will struggle to keep up with the speed at which the agents are running at.

Shan Moorthy:
So I think we require humans on the loop, which turns out is exactly what we do when it comes to management anyway. Like, I don’t micromanage my people and get involved in everything that they do. I oversee them as a manager and they come to me when, you know, they need help or I need to intervene. And it needs to be that same relationship that people have with their AI agents as well is where they delegate a task, give it autonomy to run, and they sit on the loop observing, ready to intervene, or just gathering the output and then evaluating whether it was successful or not, whether it’s two desired parameters, et cetera, and then tuning it and reinforcing it along the way as well.

Paul Spain:
Yeah, but we have one example with my team at Gorilla where it’s a particular task that will be very common across a lot of organisations. We have help desk type requests that come in so they land in a ticketing system. One thing we recognized is that nobody particularly enjoys having to categorize an incoming request or ticket and putting some effectively becomes sort of metadata against it. But that’s actually very, very helpful to be able to understand what’s going on within a help desk and to make sure things are routed to the right people and so on. Yeah.

Shan Moorthy:
And that’s based on patterns that it’s seen before.

Paul Spain:
Yeah. So we were able to find that that was a really good use case of using AI. Very light from a computational perspective. Cost virtually, you know, nothing from, you know, overhead cost to do it and you know, an immediate sort of saving on. On the time you’ve taken that from, you know, X amount of time and. And not enjoyed time as well from a human carrying out that task.

Shan Moorthy:
So the key to that is actually not having the ticket created in the first place where people can speak to the system and the system can answer based on knowledge that it has, based on the data that it has access to, the context that it has, the permission it has access to that it’s been granted. And so there’s a couple of things there. I’ll start with the example. The first one is we have hospitals that use Workday and one of the biggest problems that they have recruiting nurses is that nurses work shifts.

Paul Spain:
Yes.

Shan Moorthy:
And most often it’s the shift timings and HR’s timing doesn’t line up or the talent people’s timing don’t line up. So when they finish at 3am in the morning and they want to organize an interview or apply for a role, they. There’s no talent person on the other end able to schedule a meeting with them. So instead if they interact with the agent, so we have a talent agent, and if they interact with the agent, the agent can actually see the calendars of all the people involved and they can automatically offer time slots and they can organize those interviews. Not only that, it also sits in the interview as well. And then it gives them. Gives the interviewer feedback on skills that it has detected based on what has been said and. And also help them sort of shortlist candidates or not.

Shan Moorthy:
Not shortlist, but go through their resumes and say, you know, these are the

Paul Spain:
skills that I picked up from Highlight some of the.

Shan Moorthy:
Some of those things.

Paul Spain:
Yeah.

Shan Moorthy:
Which. Which kind of makes life easier for. For those people. The coming back to the experience side of things. So one of the things that we’ve discovered is this new way of working or this hybrid way of working with people and agents requires rethinking the way you work. So this is what I was Saying about redefining how work happens. So we have Sana. So Sana for Workday is what Gemini is to Google, but effectively what it becomes is this new front door to work.

Shan Moorthy:
So rather than having, hypothetically, rather than having an onboarding process for a person, being a. Here’s a 500 page manual. Learn everything and retain it in the back of your mind as you go through your daily tasks. What if it was just go to the page and just say, I want to do this. And it guides you through all of that using the latest version of whatever policy that applies to that, using the most accurate amount of accurate data that’s available to it, using the context and the tips that you need to get the job done faster. So that’s the hypothesis behind Sana being this unifying experience that we interact with to absolve ourselves from having to learn the screens and the navigations that we traditionally had to. Yeah, it’s this notion that AI is the new UI and not just the UI for our platform. But, you know, and we, we acknowledge that work doesn’t just happen in Workday.

Shan Moorthy:
It happens in Salesforce. It happens in all these other ancillary systems as well. So how do we actually reach into those systems and bring those components in as well? So that you can sort of. It’s not just the interaction with the agent, but also the orchestration behind it as well. So that you can have, you know, if for example, someone wants to find out what was the commission paid for someone in their team, it doesn’t. It has the context to not. To not give the commission of other teams as well.

Paul Spain:
Yeah.

Shan Moorthy:
So it’s a combination of having access to agents, the orchestration layer, and also the security and privacy models that are required to work safely as well.

Paul Spain:
Yeah. Which is super critical. Right. And I think that’s one of the areas. Sometimes organisations will get caught up in the excitement and maybe the wrong tools will get used in the wrong places where security and data privacy isn’t respected. Appropriate. And I guess we’ve sort of seen that with a range of situations and some of those things.

Shan Moorthy:
There’s a greater risk of that happening with general AI where the organisation has to put those frameworks together, rather than in a vertical AI where there is already strong governance and policies in place and frameworks in place. The other thing around that is the, the use of that AI as well. So it, I think last time I was here I was joking that it is becoming the new HR because they’re onboarding AI agents, training them up and Then you know, putting them in the workplace. But it is also being tasked with being the manager of all these AI agents as well. So how do they increase visibility? How do they manage permissions? Now in the market there are HRIS systems, HR management systems and there is ITSM systems, IT Software management systems. But AI agents aren’t necessarily software. And traditionally what you do is you onboard a piece of software, you test 10% of it, you guarantee that the other 90% works the same. You give it an admin account and it has admin level access to execute this task.

Shan Moorthy:
Right. And you might have role based access controls and things like that behind it, but software is managed in a certain way. Whereas AI agents, based on who is calling it and when they call it and for what task might have particular permissions, that is more like an employee rather than software. So what you end up with is a requirement that you need to manage AI agents in a way that is unique. You can’t throw an ITSM system at it. You can’t throw an HRIS system at it either. So it has to be somewhere in the middle. It has components of software management, it has components of HR management as well.

Shan Moorthy:
So that’s where the way we look at it in Workday is that it is this hybrid model. It is an extension of your workforce, but it operates differently to how people operate, obviously. And so we built this agent system of record which allows organisations to manage the security and permissions. It allows them to track how it’s being consumed, who it’s being consumed by. And one of the things that we, we hear from a lot of the early adopters is they want to know exactly where they’re spending their money on the AI as well. It’s very tempting to spend a lot of money on AI.

Paul Spain:
Yeah.

Shan Moorthy:
So a lot of them having been bitten by that, remember what I said, 5% proof of concept gets to production. They want to make sure that they’re measuring that ROI as well. So it almost becomes a requirement that if you want to go from experiment to production, you also have to have the ROI that goes with it and to be able to prove that. And that’s one of the things that we do with the agent system on record, is no matter who provides the agent, whether it’s our own agents, agents from other software vendors, or even if it’s agents developed in house by the organisation, you’re able to track not only what has it done, but also how much is it costing you so that you can, you can work out the ROI and you can decide whether you want to do more with it or whether you should do less with it.

Paul Spain:
Yeah, gotcha. Okay, that’s really important. Now a lot of organisations will, you know, will be using, for instance, Microsoft Entourage for, for their identities. Maybe they’re dealing with something else, Google ecosystem, et cetera. How do you connect into those platforms? I think Microsoft have recently started with entre ID identities for agents for agentic AI.

Shan Moorthy:
So the announcement from Microsoft was that they will use us as a canonical source for AI agents. So if you onboard an AI agent into Workday, it will automatically get an entra ID from Microsoft and then you can manage it that way. So there is a very close partnership that we have with Microsoft and that’s one of the outcomes of that, is that. And the other one is any agent that you build in copilot can automatically be enrolled into the agent system of record as well. So we have published on GitHub a public spec, so anyone can see it and basically is an extension of the A2A standard, the agent to agent standard, which allows agent developers to also include a certain amount of extra metadata that allows us to track what it’s costing and how it’s working and things like that. So there is a agent partner network that we have, which includes Salesforce, Adobe, Microsoft, aws, Google, and you know, these companies that most people are using to build their AI agents. And if you use any of them, there’s, it’s, it’s seamless to. To onboard that agent into Workday as a agentic teammate so that you can, you can see the people working alongside AI agents as well.

Paul Spain:
I remember we were chatting last year, there was, I guess there was a lot of sort of personification of AI agents. Is that sort of still the approach or are we thinking slightly different about agents? Because I don’t know, to me it felt a little bit uncomfortable that we’re like thinking of AI agents through such a human lens. How has that sort of evolved?

Shan Moorthy:
Whenever I come across that question, I always go back to the old adage for cloud technology. Name your servers, just don’t treat them as pets. If we take the analogy, we went from metal boxes that we treated with, we worshiped them and we protected them with our lives. And then when we went to the cloud, we literally virtualize them and throw them up in the cloud and we still kept protecting them with our lives, not realizing that if we treat them as disposable and we use the weaknesses of the cloud as its strengths, the elasticity of it, the resilience and all these things that we can have servers that scale out and scale down as we need. Right?

Paul Spain:
Yeah.

Shan Moorthy:
That means we needed to be less precious about our service, but treat them in a different way. And I think the same thing has to happen with AI agents as well. If you name your AI agent, yes, it does give you a personality. It does, I think culturally it allows you to become familiar with the AI agent, but it also means that if the agent doesn’t perform as well, do you want the, do you want to fire Jarvis or kill it?

Paul Spain:
Right. Because you might be turning these things off.

Shan Moorthy:
Well, it’s constantly evolving. You might be building another agent to take that agent based on a new model or a new work paradigm, a new way of doing the same task. So being able to swap those out more readily I think is culturally one thing we need to be okay with. The other thing is AI agents don’t necessarily fit into the org chart as easily as everyone assumes. Right. Because the same agent can be in multiple teams doing slightly different tasks as well, based on what it has delegations to and so on. You could pick up an agent and drop it in another team and it might pick up that team’s authorisations and do something completely differently. So it’s no longer a tree based org chart.

Shan Moorthy:
That comes to mind when you think of AI agents as well. So the same AI agent with potentially the same name might be doing two completely different tasks. Because the other thing that we do with AI agents is we think of them as having skills exactly the same way that you onboard a person and they might develop new skills on the job and so on. You onboard an agentic teammate and it might have a skill set that it has right now, but we’ll be constantly adding to that skill set over time as well. And that might make it more appropriate for different roles and different jobs as well. So the way you conceptually think of an AI agent, this is what I mean. It’s not quite human, but it’s also not quite a piece of software.

Paul Spain:
Yeah. Yeah. Okay. Okay, that’s good. What, what are the things that

Shan Moorthy:
you,

Paul Spain:
you’ve seen from AI gentic AI, you know, over the last last six months or, or so that really stand out as, as great examples and great use cases within organisations that get you excited

Shan Moorthy:
around

Paul Spain:
the benefit for organisations that are using them.

Shan Moorthy:
We’ve seen organisations that would traditionally have to scale up and down. So one of the examples is in the fast food industry.

Paul Spain:
So when you scaling up and down, hiring and firing sort of, and it’s

Shan Moorthy:
because it’s seasonal, because, you know, they might need to. So there, there are organisations where they need to go through 500 candidates per day, which means in order to deal with that, they have to scale up their recruitment team during, you know, the, the on season and then during the down season. They don’t need as big, they don’t need to carry as big a team. What they’ve been able to achieve with the AI agents is that they can carry a relatively static team and deal with the spikes because the spikes are automatable, or the way you respond to those spikes can be automated.

Paul Spain:
Interesting.

Shan Moorthy:
Which has saved them millions of dollars and more. More importantly, it has, it’s actually freed up the recruitment team to do more valuable tasks as well. And that’s something that we’re seeing across a lot of organisations, is most people come to work and they don’t get through the nine to five, having ticked all the boxes and done everything that they were supposed to do. They usually come back the next day with more stuff from yesterday. And that can lead to burnout, it can lead to cognitive overload and so on. What we’re seeing is by applying these AI agents, they’re able to allow people to have satisfaction in having completed things. And it also buys them some capacity to do the creative thinking. Because one of the things that we need organisations to do is to fundamentally redesign processes, because all processes to date have been built around human execution and with limits and even rules and frameworks put around them around human error rates and so on.

Shan Moorthy:
When we get to the agentic world, there are novel ways of getting things done that require experts to fundamentally reimagine how that needs to get done. And they won’t get to that unless they are given time and space to do that. So the biggest benefit that we’re seeing a lot of organisations achieve is giving their people time to actually reimagine how work gets done so that it can be done faster and safer.

Paul Spain:
Okay, yeah, that’s good. So what does that typically look like? And you were sort of talking earlier

Shan Moorthy:
around

Paul Spain:
AI having or, sorry, IT teams having responsibility across agents. But of course, the IT team doesn’t necessarily know, certainly from the outset, everything that’s involved in everybody’s job. So what are you seeing the collaboration looks like and what works in terms of that, that sort of rethinking and that collaboration to, you know, to deliver useful AI.

Shan Moorthy:
So I do joke about the fact that it is becoming the new hr, but there is a, there is a thread of truth to that in that the way HR works is they onboard a person into the organisation, they teach them the rules of the organisation and then they hand them over to the, the business unit where they are supposed to be effective.

Paul Spain:
Yeah.

Shan Moorthy:
And they help manage that person. You know, their, their how they grow or you know, how if they need to be performance managed and so on. The role of IT is becoming that they onboard an AI agent into the organisation. They have to give it some guardrails, they need to put it into a ecosystem where there are certain security and privacy and policies applied to it. But then they do need to hand it over to the business unit who is going to be responsible for training and reinforcing that agent as well.

Paul Spain:
Yes.

Shan Moorthy:
So it’s the old paradigm of a centrally managed IT service doesn’t quite work in isolation anymore. So it needs to be a partnership. And what we’re seeing in our platform, we’re seeing a partnership with IT and HR or IT and finance being the most productive ones. It has got to a point. So there’s, there’s an article online about a company called Medina Moderna Pharmaceutical company and they have appointed a chief work officer and it’s this. And they sit between IT and HR and they’re responsible for work. It doesn’t speak to who is doing the work, it just talks to the capability of doing work.

Paul Spain:
Gotcha. So it could be technology doing the work, it could be people doing the work.

Shan Moorthy:
So what they’re doing is they’re crafting how that work gets done and all the things around that and then they get that job done using a combination of the two. And I think the future is going to be IT working collaboratively with hr, working collaboratively with finance to define what work means and then jointly fulfilling. Who does that work as well?

Paul Spain:
Yeah. Now back to what you were sharing around reducing those peaks and troughs. Have you, have you got some practical examples you can share?

Shan Moorthy:
So one of the organisations that we work with had to process legal documents, contracts, for example. Now what would have traditionally taken them thousands of hours of interns and going through various documents, filtering that out, and then having legal practitioners go through and then translating them and finding the relevant information and so on, has now been condensed to a few hours of work by an AI agent that’s then delivered to an SME who’s able to take that information and act on it.

Paul Spain:
Right, Right. Okay. So yeah, the AI means at that particular busy time, you’re not having to be. Everyone’s not as stretched, but ultimately still the subject matter expert gets to play their part, but they’re getting that assistance. So the workload’s not as crazy.

Shan Moorthy:
Yeah. And it means that you don’t have to based on what’s come through the door. You don’t have your entire workforce swinging towards one thing and then everything else backing up and so on. So that whole everything has to be linear and has to be done in a synchronous manner kind of starts breaking down. So now you got organisations which can actually start thinking asynchronously. And we’re seeing this in software engineering as well, where we have engineers who can delegate a task to an AI agent, go to sleep, get up in the morning, and the AI agent has found 30 different ways of doing the same outcome and then they evaluate which one is the right way of doing it. They’re applying their SME knowledge. And I think that’s what we’re going to see with AI agents as well, is where we delegate tasks to AI agents, they go away, they rationalize it and they come back with a solution.

Shan Moorthy:
And there is still need for an SME to evaluate the outcome of that AI agent.

Paul Spain:
Look, look, looking at, I guess the, you know, the, the challenges of getting from organisations with no AI or sort of, you know, casual interactions with, with the likes of, you know, a chatbot, you know, co pilot chat, GPT, you know, Gemini, etc, from, from that, you know, to an organisation that is leveraging, you know, AI well and appropriately across, across the workforce, what can you share in terms of, you know, what, what you’ve seen that, you know, really, really helps to get to, you know, to get to that, you know, end where it’s not just kind of ad hoc casual AI usage, but it’s starting to get well implemented in the right parts of an organisation where AI can help and done in a manner where there’s confidence from a trust perspective, governance perspective, and you’ve actually got a happy workforce that are feeling and seeing the benefits without feeling as though these things are being done so that they can be left on the trash heap sort of thing. Right?

Shan Moorthy:
So every organisation, a company, so individuals at home using it, you can use it, you can use ChatGPT, you can use Gemini, for example. But a company using it has to look at it through a slightly different lens. In Australia and New Zealand, we always talk about governance risk and compliance. The GRC framework, right. So every organisation has a responsibility to make sure that those GRC frameworks are there and that it applies to people. So it has to apply to the AI agents as well. So any AI that is used, whether it’s generative AI or let’s even machine learning, generative AI or agentic AI, all has to go through that governance, risk and compliance framework, which means it needs to have the right guardrails, that. Right.

Shan Moorthy:
Deterministic components in place to ensure that the explainability is there, that the trust is there, that the observability. And it’s not just a small sample or a subset of operations, it’s 100% observability of what’s going on. Because AI, in being probabilistic means that just because it does something this way now, tomorrow won’t do something different.

Paul Spain:
Right.

Shan Moorthy:
So yeah, having those guardrails in place is, is an absolute must. Now the, the inverse of that, and, and this is the one that I always caution leaders about, is the governance, risk and compliance frameworks in most organisations are currently articulated to deal with human risk or risk that humans present to the organisations. So for example, a salary review by a person, they might get through 20 people in their team in a day. If they get that calculation wrong, that’s maybe 20 apologies that they need to send out. However, if an AI agent is given the same task, you could do your entire organisation in five minutes. Now, if that calculation is wrong, that’s on the front page of the news.

Paul Spain:
Yeah, yeah.

Shan Moorthy:
So it’s the same operation, but because it’s done at a different scale, it presents a very different risk posture. So one of the things that I encourage a lot of organisations to do is to fundamentally go back and look at their GRC frameworks as it’s articulated right now, because it’s, there’s a very good chance it’s only articulated to deal with human risk. If you apply an AI agent on top of that framework, one of two things will happen. Either the AI agent will be crippled and it will only perform at human speed and scale because that’s all your framework allows it to do, or it will continue to operate at agentic scale, you just won’t know it is. Which is, I don’t know, which is scary. Right. You pay a lot of money for an agent that operates at a human speed and scale, or you have an AI agent that’s operating at a speed and scale that your risk framework can’t even comprehend.

Paul Spain:
Right.

Shan Moorthy:
Yeah, both of those are bad outcomes. So that’s why not only is it important to have a platform like Workday that gives you the right guardrails, but also having the governance, risk and compliance frameworks in the organisation that allows AI agents to Thrive.

Paul Spain:
And I guess this is the challenge because in a tech world we often talk about doing things at pace. Zuckerberg sort of spoke about moving fast and breaking things. You can’t actually do that in every context. I think it was, there was a company down under got some attention last year for some content that was AI generated and it ended up creating a bit of a problem because they seemed to lack the appropriate governance and process wrapping around a report that they delivered. That sort of thing could happen to anyone. So there’s kind of a journey of maturing these, these policies and then how they actually operate within each organisation.

Shan Moorthy:
The great thing about the cloud revolution is that it actually taught us a lot of valuable lessons. So one of the big lessons that we learned from adopting cloud is that you have to ring fence the risk.

Paul Spain:
Yes.

Shan Moorthy:
So you have to give, going back to cloud, engineers should be given the freedom to deploy. But don’t deploy everything all at once. Deploy one small thing, send a canary deployment out and then if that works, then send the bigger production deployment out. These are new ways of working that we learned along the way. Because if you did something, if it’s a binary thing, it could be catastrophic. So I think the same thing applies with AI agents and AI in general as well is as you adopt it, ring fence a risk so that you are, it’s culturally, from a cultural standpoint, people are familiar with the thing. Trust is easy to lose, it’s hard to gain. Right.

Shan Moorthy:
So as you go through that, you, you need to earn the trust of the organisation, the business units, the people that are going to be using these things as well. So as you, if you ring fence it, it’s a lot easier to, to control and it’s also easy to champion as well. So. But the, the big thing is going from experiments to production and that’s something that I think organisations need to, I think it’s also because, especially in Anz, Australia, New Zealand, we’ve kind of become this massive transformation wave kind of way of working. I know we’ve all gone down that agile at scale path, but it’s just smaller waves of transformation. You know, we had the digital transformation, the big data transformation, the cloud transformation, mobile transformation. You know, it’s, everything’s a transformation.

Paul Spain:
Yeah, yeah.

Shan Moorthy:
I don’t think we can do an AI transformation. Right. And that’s, and I think that’s what’s catching a lot of people. You know, going from that experiment to production side of things is I think there’s a real requirement to do continuous innovation. So make Small bets.

Paul Spain:
I agree. Yeah.

Shan Moorthy:
And ring fence at risk. Prove that it works and then go bigger.

Paul Spain:
Yeah. And look. Yeah, nothing’s ever finished either. When it comes to the life of an organisation and its technology should be the same, you don’t just, oh, we’re going to do this one big thing and it’s all done. Because the needs of every organisation will keep changing. And I think even that cost of trying to do the one and done type project, often they fail. Right. When you try and do those things really big.

Paul Spain:
I guess when things do go wrong, which they will. Right. And I’ve mentioned one company, but these things will be happening right across every organisation will have some stumbles and falls and those things actually end up becoming, you know, becoming great learning. Although there’s a bit of pain when they, you know, when they happen, particularly if it gets seen, you know, externally like that one. But, you know, the ultimate result is we all learn from that. You know, when someone makes a mistake in public, that then, you know, helps us, you know, think around what can we take away.

Shan Moorthy:
One of the, the biggest learnings that came out of this is this. Well, not that, but that whole deterministic component of it. If you do everything probabilistically, it’s a recipe for disaster. You absolutely need that deterministic side of things to balance out the opportunity that the probabilistic side presents. And that’s we’ve had. You know, every single organisation has a choice. They can stand on the shoulders of platforms like us who provide some of those rails, or they can build their own ERP from scratch. And really, are they in the business of building ERPs or are they in the business of training students or, you know, making food or what is their actual business goal? Sure.

Shan Moorthy:
So those are the things that we’re seeing, you know, catch people out. There’s also a cultural component to it as well. You know, when we started the conversation I mentioned. So, for example, about a year ago, if you said give autonomy to the agent to figure out the work and come back to you with a result, you would have called me a heretic. Whereas nowadays there was a moment actually in our timeline where OpenClaw came out and suddenly it became culturally okay to give your admin access to your machine, to an AI agent to do all sorts of weird things about. Now, that was an over index. But I think it then became okay to give it a level of autonomy. And so culturally as a population, we kind of went, okay, we can give AI agents some autonomy to figure out what to do.

Shan Moorthy:
And then it has some benefits to that. But I think there just needs to be a counterbalance of do we have the right checks and, you know.

Paul Spain:
Yeah, yeah, I mean, I see that. Yeah, that’s, that’s still being worked through for sure. I can’t remember who it was, but I saw I was looking at somebody’s profile on LinkedIn who sits within Microsoft and I hadn’t realized Microsoft were formally doing anything with openclaw. But yeah, this particular person’s role, that was, that was part of what they were, you know, what they were doing. And yeah, I think we had a pretty interesting podcast which, you know, those didn’t catch. It can go back and listen into probably a couple of months ago, sort of, you know, talking about OpenClaw and looking at it from some differing perspectives. But yeah, lots of takeaways when we get something, you know, that is so different to what we’ve, what we’ve had in the past and it challenges a lot of our existing thinking and yeah, encourages us to step back and try and figure out what is the best way to move forward with this technology.

Shan Moorthy:
So in the same way we’ve had organisations come to us and say, we want to build AI agents, but we don’t want to solve the privacy and security concerns. We want you to do that. How do we do that? And we don’t have the experts in. And this is the thing, the organisations like Microsoft, for example, do have the experts in house to build agents from scratch. We have the experts in house to build agents from scratch, but a lot of organisations don’t and they’re trying to figure out how to actually build those agents but still not have to do all the low level work that’s required to build them safely. So we’ve got the Workday build as a platform which allows you to build AI agents. We acquired an open source platform called flowise and we’ve incorporated that with Sana and another acquisition called Pipedream. So Pipedream allows you to connect to 3,000 different applications.

Shan Moorthy:
So it absolves you of all the integration problems. Right.

Paul Spain:
Super helpful.

Shan Moorthy:
Yeah. So now you’ve got an agentic way, Sana being the way that you interact with the AI agent in an agentic native way. Agent, native way. And then you got Flowwise, which is a agent builder, open source. And then you got pipedream, which has all the connectors, which instantly gives organisations access to all of this that sits on top of our data model and our security model so they don’t have to reinvent that which means they can focus on these agents doing things that are novel without having to solve everything from zero. And as soon as you’re done building those AI agents, they automatically get enrolled in the agent system of record. They, they can be deployed to their whole workforce. It just works seamlessly.

Shan Moorthy:
So and Flow Wise is still open source. We will keep it open source and you know, people can go in the community. There’s a thriving developer community around it as well.

Paul Spain:
Great.

Shan Moorthy:
And you know, they can use it for all sorts of things.

Paul Spain:
Yeah, yeah. Oh, fantastic. That’s really good. Yeah, it was interesting last week Microsoft had their AI day here in Auckland and yeah, there’s quite a range of speakers from, you know, across industry and government, including the Chief Executive such in Adela. And yeah, there was some pretty fascinating discussions there and I think some of that content is available online now. One of the couple of interesting bits that sort of jumped out was along your lines, what you’re talking about, about the proof of concepts. Yeah, it seems lots of people have tried proof of concepts, but yeah, some of the stories we were hearing, yeah, some organisations haven’t moved much further than that. They’re probably still, you know, stuck a bit there trying to, trying to work out what, you know, what are the best use cases and, you know, how do they move forward.

Paul Spain:
Another example, one of the speakers I think from within Microsoft was talking about developing an age and I think for her boss that would, you know, take actions on similar types of emails that that person gets asking for, you know, guidance on a particular thing and realized, well, actually we can use an agent to, you know, figure out what, what they’re asking for and fire some information back. But as with all these things, there’s, there’s a level of testing. This thing went, went live. It was not set up to just deal with the new incoming messages, but she, she noticed it was triggering on, you know, email messages maybe a decade old and trying to fire back, you know, answers and guidance there. So yeah, just, I think it’s just good to, you know, keep the learning coming on that front. We’re chatting here Tuesday 28th of April, big Workday event in Auckland tomorrow. Tell us a little bit about that. Now, I guess the timing is such that if folks aren’t registered, they’re probably not gonna make it.

Paul Spain:
But I imagine there’ll be online content and so on available.

Shan Moorthy:
There’ll definitely be online content, content as well. So Workday elevate is the big event. We have a innovation and main stage presentations on all the new cool features that we’ve got, but we’ve also got hr, finance and technology tracks as well. So for people in the HR field they can go and see what’s relevant in that track. I’ll be heading up the technology track. So we’re going to get super geeky and we can go into the architecture of how to build an agent and there’s even hands on labs where you can try out the new things and see what’s possible. And really if you’re used to building agents using a general AI platform like Copilot and so on, definitely come along and have a look at what it looks like in a vertical AI platform which has a lot of those common sense guardrails and everything in place. And I think that really will be the boost that you need to go from experimentation to production.

Shan Moorthy:
A lot of the time that too hard basket is always the checks and balances that are the mundane stuff that you have to solve which stops engineering teams dead in their tracks and takes the wind out of their sails. So if you see it in action in a platform like Workday, then I think there’s a very high, a much higher chance that you will take that proof of concept to production. So come along, come to the technology track. I can go as geeky as anyone wants to.

Paul Spain:
Excellent, excellent. That’s good. So that’s at New Zealand International Convention centre, the same venue Microsoft were using, one of the first, their new convention centre. So that’s exciting. Yeah, looks like a really impressive event. For folks that have missed it, where would they look to catch up with content that’s either been there or one of your other elevate events?

Shan Moorthy:
So if you do register for the Elevate event, even after the event is finished, we will have the online version of that content made available as well.

Paul Spain:
Oh, brilliant. Okay, fantastic. That’s good. I think we’re about out of time. Anything else that you wanted to add, Shan?

Shan Moorthy:
Look, it’s, I’m sure I’ll be back here in a year’s time and we’ll be talking about, you know, how AI seems to be going at a mind boggling pace and maybe we’ll be talking about AGI, you know, the, the general intelligence, artificial General Intelligence next year. But in the meantime, look, I think a lot of organisations are also having FOMO. They’re looking at the big conferences and seeing big organisations talking about a big game about how they’re leveraging AI and being super productive. I think there’s a fear of missing out and there’s also a Fear of, where do I start? The biggest temptation is to grab the nearest tool and try to build it all yourself. And I think the best thing right now is to settle on a platform that services your needs right now, but also gives you the pathway to expand in the future when your organisation is ready to step into that space, make those critical decisions. Now, a platform like Workday, for example,

Paul Spain:
you would mention Workday, of course.

Shan Moorthy:
Well, if you don’t want to use AI, you can just use it to manage your people and your money. Right? But when you are ready to manage your AI as well, it’s there and it’s safe and it’s trustable. So partner with a platform that will grow with you and make small bets. Build that trust over time and leverage, leverage, leverage, leverage. The world is moving at a extraordinary pace that if you try to do everything yourself while you learn to walk, everyone will start to run. So by leveraging, you can stay ahead of the crowd or stay with the crowd at the very least.

Paul Spain:
Yeah, that’s good advice. And I think it is important to be strategic with this, the selection of the tools and technologies because they play in over time. And if you’ve got people grabbing all sorts of bits and pieces, that’s going to get pretty painful at some point down the track if you’re not planning. And it becomes probably in some cases almost impossible to put guardrails in with.

Shan Moorthy:
You know, it’s harder to reverse engineer the guardrails than having a system with guardrails built in as, as part of its core foundation. And the other thing is, I think it’s inevitable that the future is going to be a hybrid workforce of people and AI agents. You can’t deny that anymore. I think that ship has sailed. So plan for that future, if not today, tomorrow.

Paul Spain:
Yeah, yeah, good stuff. Oh, well, thank you so much and we’ll, we’ll look forward to. Look forward to the next chat. Thanks everyone for listening in and of course, big thank you to our show Partners, Gorilla Technology, Spark, 2degrees, One NZ, Workday and Fortinet. Really appreciate that support and great to have Workday represented today by Shan. So, yeah, fantastic. We’ll catch everybody again next week. Thanks, Shan.