Future Trends in Physical AI

For thirty years now, running a company meant running applications. You logged into a screen, you typed into fields, and the software did exactly what you told it. My guest today has spent almost 25 years building that software and he thinks that era is about to end. We're shifting from applications to agentic agents. So so let's get into it. Welcome to FutureBytes. I'm Magnus Oxenwald, VP for Group AI at Columbus. And on this show, we cut through the AI hype and into the real story. What's uh actually working, what's failing, and what it means for the people who have to make it work. My guest today has been on the front row seat for a quarter of a century. That sounds so long, but it is. Ving Nash Subram ani has been within for almost 25 years, and he leads the technology that runs across the entire company. It's very broad. He does AI, he does data , automation, analytics, governance, the entire stack. He's led AI there since 2017 , and he's driven in for agentic AI strategy since the day chat GPT landed. He also sits in the room with teams building other frontier solutions with Amazon Anthropic, many other companies having this dialogue. So Vignesh, welcome to Future Byt es. Thanks, Magnus. Thanks for having me. Really excited and looking forward to the talk today. Me too. This is gonna be amazing. I wanted the listener to know more about you because you're a truly interesting person . And if you could tell us your journey, break it down uh before we get into the um the actual talk. Absolutely. So this August I would be completing 25 years at In for . When I joined the company, it was the days of Bonn ERP. I had just graduated from college with a couple of years of experience, and I saw this nice advertisement in the newspaper that said, Do you want to come and live and work in the Netherlands? And I applied for the job of a QA analyst . That time I got through the interview, came to the Netherlands, you know, been through the entire journey of uh enterprise applications evolving from the two-tier architecture, client server architecture to the middle tier, the e-commerce days, the mobile revolution, the SARS journey, and now the AI era with the Asian TIC. So frankly, it's been never been a dull moment, always been at the forefront of evolving technology changes. And before I blinked, it's uh 25 years, and here I am wow but it's 25 years that that's rare yeah you you stay there for twenty five years absolutely I mean uh of course it's a great company to work for which is a big reason why I stayed, but also the reasons that I mentioned, we get to operate in this startup mode perpetually. We are always on the bleeding edge. We are not following the trend, but we are in many ways on that layer which defines the trend , and the at least the pioneers are the initial group of people who lead the trend. That's been representative of my journey at Infor all along. Hence one of the biggest motivations for me to stick around so during this journey, I mean so much things has happened. Uh, we're we're targeting you know this AI shift from apps to agents, and I'm just curious , do you have like a specific moment where it stopped feeling like just another technology? Like it's just cloud was a big leap for sure, but this started to feel like something bigger. Correct. What was that for you? Yeah, it's it's very difficult to pinpoint a single moment, but clearly the launch of Chat GPT November twenty twenty two, I believe , can be identified as one of those moons. Basically, it's it's a slow burn that has been happening for some time, which suddenly ignited around that time frame. Let's say before that, we were busy with the prescriptive and predictive machine learning based AI, right? We used to do a predictive analytics here, anomaly deduction there. But around that 20 22, beginning of 2023 timeframe, the frontier models came into the picture. They started demonstrating genuine reasoning, not just pattern matching. I realized then we are not talking anymore about a better feature , but we are talking about a different architecture of how software works, right? When you start seeing enterprise applications not just automating or uh executing a single task based on intelligence, you see this multi-step business workflows which a front which a frontier model could navigate end-to-end uh without having to explicitly telling the model what to do without explicitly programming for every branch that it could take. That's when I thought this changes the application layer completely. We are not uh adding intelligence to the software, but we are building software around the intelligence. Yeah. Yeah, we're gonna get into more details later, but uh before before we get in there, can you share your feeling around this? I mean, you sit in in Info, you have a lot of internal conversations, but you also link to a lot of other big companies and their development processes. What in these rooms generally excites you right now, but also what might keep you up at night? What are the top things like I think the answer is the same for both the question, which is the breakneck pace, the speed. The speed. Things are happening right now. It's mind-boggling. Every two months, every three months, there is a earth shattering development , which you realize is going to have material consequences around the way how we build software, how we deploy, how we support everything. Um, so obviously work ing with the other technology pioneers in the industry, uh what we see now as the trend to watch out for is the the long-range agents, particularly around February this year, these long range agents came into the picture, like the cloud co-works, um, and the open open AIS codex. Um, you know, cursor is another tool in this framework, and so on. So these long-range agents uh are able to plan and executa over a very long horizon with minimal human supervision, not just answering questions, running sequences of actions over time with judgment, when to proceed, when to stop, etc. I don't think everybody's really understanding the evolution around this long-range agent agents. That's number one. And secondly, you know, the tools of the trade of these agents are also maturing immensely. Mainly, I'm talking about things like the context window and memory. These are all, let's say, the runtime infrastructure under which these models and agents operate. Those are just exploding, the supporting more and more capabilities by the day. So when a model can actually hold the entire business context like your customers' history, your vendor contracts, your inventory positions, all within the same context. The quality of its judgment changes dramatically and accurately, right? So then it's not a feature. We are talking about it. This is a different kind of software. Yeah. And while you're running and somebody is now telling you run even faster, because it this is not a linear development, right? It's like uh it's exponential, it's moving faster and faster. At the same time, you need to have one foot in in force business uh model and requirements of what you need to solve, what kind of problems you need to solve for your customers and value you deliver to your customers. And at the same time, the other foot, you need to be an expert in the evolution and the development of AI because you need to match those opportunities right like you say how does a uh uh a growing context window affect your business model uh at inf o how do you find the time what what what kind of organization do you have around yourself to actually find time. Or do you never sleep? No, I mean uh we need to stay sharp. We do have um lot of tools at our disposal. We are part of industry councils. We are part of product advisory boards with our customers. So we get to directly hear from our pioneering customers what their expectations are. We speak a lot with the analyst and the analyst communities day in and day out. Uh of course even tools like LinkedIn are a fantastic resource to just uh be on top of everything that's happening. So yeah, you having all the developments happening is one thing, but being on top of it and having a constant plan to respond, that's indeed uh not everybody's cup of tea. Yeah, so so let's go back and double click on the actual shift. So you said something very valid, people haven't truly understood this change, right? So in your own words , what is changing right now that matters most for business leaders to understand how to adjust their roadmap. Obviously they can just buy in four products and hope you're doing the right thing, but if they want to get involved and understand what you're seeing, how would you explain that to them? Yeah, I mean the question that every business leader should be asking themselves right now is which parts of my business and my operation am I comfortable um being run by the agents while I am not watching or with minimal oversight, right? So that is the biggest judgment which a business leader needs to make. Uh, because we are seeing this varied degrees of autonomous execution, right? So still we come from the world of digital assistants and co-pilots, where the human is still very much in the driving seat and uses the technology for assisting themselves. And then there is the approval g ated tasks where the AI does the analytical heavy lifting, and the human then uh you know wits and follows through on what the AI is proposing. And lastly the fully autonomous, hands-free mode of agent tick AI, where you know the agents are completely performing um routine tasks, high volume tasks, and so on. So the for business leaders, I think making this segregation is uh probably the most hardest thing and the most important thing to do. We had a chat a few weeks back, and you you said even though you have this long-term autonomicity, yeah. Um the AI today is sort of jagged, it's not perfect, right? It doesn't get everything right. So if if 75% is right all the time and the rest is not, that's also sort of a deal breaker, right? Because you talked about flows and matching those flows to autonomicities isn't always uh perfect in this jagged intelligence world. If if AI is writing a first draft of an white paper that you asked it to write with eighty five percent accuracy, that's brilliant output transcript: That's fantastic. But if the AI is only 85% accurate in fulfilling your invoices and accounts payable and accounts receivable automation, yeah, that's that's not fantastic . That's an accounting scandal. If fifteen percent of payments in your company is not happening uh like how it should, that's simply not how fifteen percent of my salary. So it's very important to understand that accuracy and trust and margin of error are extremely important when you entrust an agent with a task. Having said that, you can even go one step further. Infor is big in uh food and beverages, for instance, and many of our customers are into the business of uh making animal feed, for example. And these feed gets made on time with the right amount of proportionate ingredients and so on. And all these farm animals need to be fed at a specific routine. And imagine that with 85% accuracy, if you are managing this process, the consequence could be that the animals are not getting their feed on time or with wrong concentration of ingredients, which could be catastrophic. So don't want to be interesting AI to be doing these types of mission critical tasks yet, while it remains extremely efficient and brilliant on all the uh you know routine tasks, uh, also subjective tasks that requires a lot of input, that requires a bit of judgment on the go, a little bit of analytical quotient, AI really exc els in all of that, but for mission critical tasks, we still need to have the human um as part of the equation. Now, is there an accountability factor here? So if I use a tool or a system and I buy that service from infirm expecting value, where does your accountability as a vendor start and where does it start with the with the customers using the product? Yeah. Great question. Honestly, these are the questions where the answer is not crystal clear yet in the industry. We are all trying to figure out the answer for these types of emerging challenges as we speak, uh, I think the answer is somewhat similar to how the product liability law is currently working, right? So basically the model provider is responsible for making sure the models work as documented. Yeah, the platform provider, like Infor in this case, is responsible for making sure we have grounded that model correctly, given it the right God rails, and given it the right context, um, the data access, the approval action set, that's our responsibility. Clearly, the customer is responsible for how they use it, right? I mean, if a car is capable of uh reaching a maximum speed of uh 250 kilometers per hour, and if a customer drives that at let's say maximum speed on a road where only the maximum limit is 100 and crashes the car, yeah, is that a problem of the manufacturer of the car or is it the problem of the user? So I think the biggest liability here lies with uh the user. Uh because uh they are responsible for what goal they are giving the agent, what approval thresholds they set, uh they the oversight they choose to have uh don't exactly the the dangerous shown here is that this is a legal situation and la wyers will get involved. Um a customer who might believe that the vendor is fully responsible for the agentic behavior and the vendor believes, you know, they have disclaimers and liability in the terms of service . And that's where it gets tricky. And frankly, the legal frameworks are catching up. The uh the sensitive part here is how can we keep the innovation going by these factors not really uh stopping you on your tracks. We work very closely with our legal team and QRA team on all these topics. We are very transparent on how the customers' data is treated, how the agents make their decision, what part of liability lies with info, and what part lies with the customer and with our upstream technology providers. All the communication and transparency is the key. We have to make all of this crystal clear with the information we have and the frameworks are catching up so it's an evolving area. Do you think it will be like a one model that fits everyone or will it be like uh somebody wants a more locked down service and with more guardrails and they can get that and somebody else's wants to be more like uh testing the car out and really uh driving it 200 miles per hour if it's see if it's possible . Yeah it it in a way goes back to your earlier question on the jagged nature of the AI in a way. The AI in its current form is uniformly better than humans in some areas, uniformly pathetically poorer than humans in some areas. Correct. So especially processing high volume patent recognition , synthesizing information across large document corpus , generating first drafts of structured output, of course, operating 24-7 without fatigue. These are all areas where the models excel. Whereas for humans, still the presence of mind or responding to a new situation on which there is no precedence. I think that's still our superpower. Because the models all sudden done needs some kind of a training. So the novel ethical judgment is as we speak, is a human superpower . Um low data situations where there has not been a lot of training data to go by and so on. That's that we then exercise our own judgment. In adversarial contexts, uh HR-related, personal related, um, ethical way of thinking, uh, right? All this is our superpower. So going back to to your question, I think um it depends on using the technology for its superpower and not letting it run areas which are inherently not in its strong um you know area of expertise. So okay, so so the biggest drivers for your customers now acquiring the intelligence-based products on top of info products . Would would you say then its volume uh use cases and its it's um it's uh efficiency use cases, it's it's where you see there's low risk for uh if failure happens it's it's low risk but you can still uh or w where would you see it um but it what I'm trying to ask is is are you predominantly seeing using AI now to cut the need for humans involved using the application or where where do you see the most value currently? Yeah. I think the honest rule of thumb is if I am comfortable being away for a week and I'm okay to learn what happened when I come back, all those tasks should be automated using agents. Right? That's a good one. If I have to explain to myself or to my boss , what went wrong, then that should never be automated with agents, or at least that's a scenario which should be semi autonomous with the human I think that's the most simplest way how I could frame it . Um, right. Um, so I mean, I will give you a simple example. Um, in the Netherlands, uh, we have the healthcare system, and I call my GP, my my regular doctor, uh, and when i call the practice if i am calling for refill of a prescription that's already provided that's a no human interface workflow for me i it just goes to a choice that I have to make. If I'm calling for refill of the medicine, I just press a button and then I you know, it gets delivered to me automatically or I can go and keep it up. I think there are a lot of scenarios like this in the shop floor where you have to replenish your inventory based on very simple, straightforward, non-risky triggers. Yeah. Or um repeat incident handling. It's an excellent example . So a situation happened, it was fixed with a very clean workflow. The same issue happens again. And you know how to deal with it. These are all use cases where all businesses are employing, you know, the cognitive power of people to do these types of simple repetitive tasks, which still has it's not like a rule-based task, but it's a business scenario. So I think these types of scenarios where there is uh you know routine reconciliations , um, pattern match uh response, uh, you know, even things like invoice matching, demand forecasting updates, uh reordering triggers, repeat incident handling, you know, tier one customer support respons es, uh again, against known issues and patterns, um, flagging up any compliance monitoring and alerting, all these can be completely agentic with to a fair degree of autonomous execution today . Wherever there is uh you know any external facing commitment above, for example, a certain dollar threshold or a high relationship um uh you know situation where you don't want to antagonize your supplier or customer. Actually, we have a customer who to send dunning letters customers who are had to pay them. Uh that could be automated through an agent, dunning agent, but they chose to review the communication which the agent is going to send to them before it is sent. Now that's more the semi-autonomous where the agent has already done the heavy lifting and then you do the final call and any fine-tunings. And obviously, any HR-related, personal related, high trust relationship situations, as I mentioned, novel ethical judgment, uh and reacting under situations with not a lot of precedents, those should be one hundred percent human-driven with no agent uh involvement. At least that's my view. No, I I think it's a very good rule of thumb. Then then obviously there is a lot of complexity attached to this. Everybody's unique and they have their different use cases and whatnot. And and and and going back to that, I know uh for for some of your products you've been very helpful. You instead of just giving a product that people can just go and do whatever they want to do, you you're helping them with uh a specific industry mold um something that that guides them to actually do something so you mentioned food and beverages you it's not like uh that solution is is fit for other industries because you you moved in the enterprise continuum closer to the customer, trying to help them with industry specific uh setup. How do you think do you think that still is going to be valid when you have agentic services and and and now when you're using like world models or or um uh state-of-the-art models.com, will you need to start producing your own fine-tuned models now where it's you know more for food and beverage or how do you see yourself getting involved there? Excellent question. I mean you kind of alluded to the answer in your question. So there are two paths to take, right? One is you give the tools to a customer. Yeah. And then say here are all the best practices on how to use the tool. Here is what the tool is capable of. Solve the problem yourself. And then they go to an SI or an uh AI company and then build solutions. That's one approach. Or um, you know, we are the industry experts in this case. We are we we have already done that research, we know exactly for this industry which areas AI works best, which areas you should not be using AI. That research and uh readiness is already coming from uh the v vendentoror , for example, I mean, if you are processing waste in a manufacturing facility, for industrial manufacturing, it could mean scrap metal. For food and beverage, it could be the yield waste, the byproducts of your production. And for healthcare industry, it could be the clinical waste. So when you use the term waste to any of these frontier models within an info context, depending on where you are applying the term, our models know already what the customer means as opposed to in your other situation. A customer would have to train their model to qualify when you encounter the term waste, don't use your general knowledge of uh repository. Instead, this is what I mean by waste, right? Uh, so the cost of getting it wrong in a generic deployment is paid by the custom er, whereas with a guided deployment, info has paid that tuition fee already, uh, you know, in the development and iteration. So for a mid-market manufacturing company, especially, you know, that doesn't have more than 50 people or so . The question isn't even guided versus non-guided. It's it's basically uh going with the uh guided uh approach. Makes a lot of sense, cuts down the risk, cuts down the cost. So it again, choosing the right technology and the right use cases, um, that is the key. And in the generic approach, you would miss that discretion. Yeah, yeah, absolutely. I see that. So so I I see the advantage there. It's more efficient approach and you get to your results uh faster. Now and that will hit their business model in terms of they will be become more efficient, they will require because they're digitalizing, they require fewer humans . Now, this will then reversely just hit your business model as well the better you deliver a product the less us ers will use it yeah so so how are aren't you like um ripping the rug out of under your own feet, sort of kind of or or are you thinking you're gonna make money from something else in the future? Or fair question. Fair question. I think uh the first response here is that we have to acknowledge the changing reality. Yeah, um, that is the way you can uh reach the outcome that that path is getting shorter faster most efficient for our customers i mean if you have to solve your business problems the way how you solve your business problems is getting more more and efficient through this agents and and whatnot. So then obviously it is um a recalibration that everybody needs to go through especially for instance if, you have a seed-based licensing model, that is not probably going to be long-term sustainable. You need to probably start thinking more on an outcome-based or a consumption-based pricing model, so that ultimately customers don't have a problem paying for the value. Even today, the seed-based licensing is a means to mon etize the value. And if that underlying paradigm is changing, then obviously the monetization mechanism will move towards where the value is, which would be the outcome-driven or a consumption -driven pricing pattern. But having said that, if you uh if your superpower in the past happens to be the friction and complexity that you generated, and that's how you manage to retain your customer. I think those days are gone. So everybody needs to readjust to the new reality. And yeah, it's a shift across the board and now there's a lot of new technology coming your way uh we talked about in the beginning of this talk and how do you enable all this because there's if you are running faster and faster and you're an expert in this, I mean I can just imagine how fast the customer needs to run in order to catch up. Uh how do how do you get them enabled? How do they get access to the uh will they buy capability by capability or or are you just uh how is your strategy around that? No I mean going back to again the guided versus unguided approach that in a way also applies here. We at In4 have a unique point of view on bringing AI to our customers. So it's not just the technology, but also the outcomes and the solutions. So it's not just giving a set of tools and asking the customer to make their own soup. We in fact prepare the soup according to their wishes from the day one, right? So it's more around the solutions for specific use cases how do you redu ce um waste how do you uh reduce your uh open payment uh uh timeline? do How you you know increase uh your bottom line? How do you um you know make sure that the efficiency output is going up while our costs are low? So these are all the problems with which we need to approach AI and not just saying here is the frontier model or here is the orchestration framework or you know, here is the uh supremely bigger context window-based framework, etc. That means nothing to the custom ers. So what they are looking for are solutions to their shop for problems. Um, and that is how I think we should approach this. That puts a lot of responsibility on top of the custom ers as well, right? They need to have a plan. Um, in in order to, you know, adjust to this reality. You're producing this big, this SARS tool, and and you have all these intelligence-based products around it and you have obviously industry standards so so they get a lot of uh business logic out of the box obviously and and and and industry but how did the close the last gap? You know, how do they understand how to reskill? Uh do they need to retrain staff, think differently, all these things in order to do their part to become in sym bios with with the uh application changes. Again, a very good question. And like the case with most difficult questions, the answers tend to be simple. Uh the answer in this case, customers have to do their preparation and need to be AI ready. And they can do that in a few uh paradigms. First of all, on the front of data, they need to ensure they have quality data. If your master data is riddled with redundancies, or if there is incomplete data for your vendors or for your customers or it first of all data quality that that's where I'm going with it. That does not need agent or AI or whatever to start doing. It is something internal. Everybody can start looking at their current state of data, can clean up. Of course you can use AI to do that. And we also have tools specifically on the infofront to do all that. So then that's the place where I would request all our customers to start , have a hard look at your data. Make sure that the data quality is top-notch so that when the AI and agents come in, they can do their job. That's number one. Number two is data governance ? What roles and responsibilities are allowed to access what type of data? Right? What are the business roles? Are the business roles clearly mapped with the capabilities within the enterprise software? How closely is your organization running with the best practices of your enterprise software vendors recommendation? All this, again, you can do some discovery. There are tools that help you to bridge this, find the gap and bridge the gap. But that's something which every customer can do today to be prepared for this. All these are foundation, having data quality, governance practices, uh, you know, business processes that align to the golden blueprint, all this, as much I know it's too much to ask for, but these are the organic improvements that need to happen first for the agents to come and work their magic. So that's one area for customers to focus. The other area obviously is upskilling the personal. That not that is not applicable just for our custom ers. That's also applicable for us, my team, frankly. And you you need to be ahead, right? You need to be advisors. You need to be even further than than your customers normally. Absolutely, absolutely. So that is purely a matter of reskilling. That's again a very cliched adage, but the person who is able to use the AI tools well will be dramatically more productive than those who don't. So obviously, you can start as small as things like prompt engineering and uh you know how to build good prompts, and we have to stop using these LLMs like a Google search. That is the biggest mistake I see everybody doing. It's a gross injustice to these frontier models. If you are just asking zero shot questions that you simply want to know more. These agents are meant to break down complex tasks, make a plan and execute and sit next to you, be your uh homie and help you achieve complexity. Digital coworker. Yeah. Exactly. So everybody should move away from the mode of using the new technology like a search engine to a problem solver. Yeah, like a companion, like you said, or somebody working side by side with you, like a digital colleague or a co-worker. Yeah. And obviously, don't trust this problem solver, whatever it is saying, but trust, but verify because exactly. So it's still jagged. So it's it's uh the accountability is with the human still, as we we said in the beginning. So I think these are all the basics which does not need any hi fi technology to come in, but everybody can start doing themselves . So we're we're getting to the end of this talk and and and uh I want to spend the last uh few moments uh thinking about into the future because you you have such a on perspective you because you this is your work. Um what do you see like is coming next because there's there's a frustration where if you were to explain the to be picture in what you can see, so you can give like a clear enough horizon for the business leaders. I mean, there's tons of things coming, but usually when I'm designing a digital transformation journey, I I know the as-is situ ation. And I I want to understand the to be picture. So if is that even possible to give such a picture, uh how should I think as a business leader now or what should I do? I mean we there is we we we need not hold back from bold ambitions and visions and uh futuristic thoughts so I can always uh share what I am person ally believing or will uh hap in my opinion what will happen in the days to come. Obviously the physical AI is a great new trend that aug urs well with the world of agents and the frontier models. And it's robotics, right? It's it's the robotics, exactly, exactly. The robotics. And that can be both in shop floor in the B2B in the business front as well as on the personal front, where the emergence of robot will help you with all your daily chores and whatnot. But specifically on the enterprise context, uh, you know, the dip the supply chain and uh warehouses, logistics, all this will dramatically change with uh increasing amount of uh physical A and robotics, for example. Robotic, you mentioned robotics, that's specifically true in areas like warehouses. Whereas autonomous vehicles, there is a no-robo there, but that just the vehicle itself is uh driving itself with all types of practical decision making while on the move, right? Yeah, yeah. Yeah, what will happen on the roads will obviously happen also on the logistics sector. Uh, that the autonomous vehicles is a huge game changer also in the business world, uh, in the shop floor, there will be things like sensors and actuat ors. So it's not just robotics when I say physical AI, it's like robots for warehouse, um, autonomous vehicles in logistics, sensors and actuators and shop floor. So the entire ecosystem of manufacturing, distribution, and healthcare will shift uh with the physical AI interwo one into day-to-day operations. I mean, uh, I know even our company I'm working for, they're delivering now the in-for uh warehouse management system solution to big companies, right? Yeah, uh in that future the WMS system would that be linked to in your head in the same package deal with robotics? So you it's not like SAS only, it's it's SAS plus robotics, it's the entire portfolio basically. One hundred percent. That's it. One hundred percent. Cool. That will happen. You're studying uh robotics as well now. So you you have another batch in yours, so it's data, AI, it's everything, and robotics. Absolutely, it's not a question of if, it's a question of when. That is clear. That is one trend. Obviously, there are other trends, like I mean, yeah, the world, in my opinion, will move to a continuous intelligence rather than a periodic intelligence in the business world. By that I mean today, you know, the quarter close, the annual budget cycle, uh, the monthly business review, this is how the business is run today. And software and solutions are all designed to cater for this periodic uh paradigm that we are currently in. Whereas the power of the agents and AI is this continuous intelligence. So I expect, it's my personal view again, I expect businesses would be more responding continuously rather than continuously in real time, rather than uh you know the periodic calendars. So so leaders will be making more agile decisions against live models, not the historical reports. That's another trend I predict. And lastly what info keeps saying as well the era of agent tick enterprise essentially the autonomous enterprise will be a genuine concept and not a fiction uh anymore and majority of operational business decisions will be made by AI systems humans will govern the outcomes set the strategy uh you know make sure that their organization has the competitive advantage. But definitely the era of agent tech enterprise is upon us. So we're coming to the end, uh Ignash, and and as a last closing question , if if from everything we talked about today, because we talked about a lot, is there a key thing we would give away to the business leaders listening to this from your story that you really want them to take and and put on their roadmap what what should that be that uh first point for me is acknowledgement of the reality this is a paradigm shift uh the i still see some sense of denial or thinking that the current systems are still extremely future-proof, or we will continue to do the things the way we do. I think that mentality, that mindset needs to go, and we need to acknowledge that the B2B world and the business enterprise software will change on its head, whether we like it or whether we acknowledge it or not. So that's number one, which is the realization and acknowledgement. If at all you there is any doubt with anyone, I think it's a good moment to reset those doubts and acknowledge the trend and the shift and the transformation is real. That's number one. Number two, what I already mentioned: prepare yourself by having a hard look internally, by improving your data quality, your governance practices, your business processes, and of course, software vendors like Infor can help you in all these discovery and optimization processes. That's number two. And number three is upskilling. We all need to become more AI native, use these tools more fluently, as I said, not just as a search, but as a critical thinker and a problem solver next to you. Uh, also don't focus more on the middle of a task-based paradigm. I mean, don't be that person who is in data entry or creating the first draft and things that can be easily automated, but be that person who owns the acco untability, trust , or uh domain expertise, which is which cannot be replaced. Uh, right. So I think people have to move from doing the core of the middle of the task to the top or the edges of the domain and realize their superpower and upskill on those areas specifically . Thank you. That was a excellent answer. So Vingash, you you have 25 years, the move from applications to agents, the 85% problem that we discussed, and why humans still matters in this process. The way the whole business software, even how it's built, is about to change like the business model from both ends. And the reckoning we don't like talking about and the hope uh on the other side of it, right? Uh, if you're listening to this and you're a leader listening to this, the takeaway isn't about just AI alone, it's this understand the shift is what what what Vignette says. Decide which decisions you actually hand over and only only reach for AI where where it generally wins. That's how you build a roadmap instead of just uh chasing the hype. Uh with that, thank you, Ving Nash. Thank you so much. Thank you. Thank you all for listening to Future Bytes. It's a great talk, Pagnas. Once again, thank you for having me. Thank you . You've been listening to Future Bytes, Bits of Business Transformation, hosted by Magnus Oxenwald, a leading expert in digital transformation and AI. Brought to you by Columbus, this podcast dives into the latest AI trends and innovations shaping the future of business. For more insights, visit us at ColumbusGlobal.com