Name: Stay Alert: 5 AI Trends that Will Impact Your Business
Uploaded: 2025-07-15T16:02:51.329Z
Duration: 45 min 8 s
Description: Stay Alert: 5 AI Trends that Will Impact Your Business

Transcript for "Stay Alert: 5 AI Trends that Will Impact Your Business": Hi. My name is Jim Little. I'm the chief innovation officer for Nasuni, and it's my great pleasure today to be joined by Rachel from AWS. Rachel, if you could take a minute to introduce yourself and maybe your background and a little bit about your role at AWS. Thanks, Jim. Happy to be here. As I said, my name is Rachel Radishzewski, and I'm with AWS. I'm responsible for our strategic partnership in the infrastructure center of excellence. Our team is responsible for setting strategy for partnerships, including artificial intelligence, generative AI, and all other partners that work along in that stack. Great. So great to have you here. We're gonna be talking about five key trends in our fireside chat today. First one's AI trust. Second one is ROI, a big topic. How do you maximize value with our with AI? Third one is model context protocol, more fondly known as MCP. Fourth one is retrieval augmented generation, or RAG. And the last one is AI agent. So really, the first two, I would say, you know, more to do with business, business trends. And the last three really a little more technical, but we'll try to elevate them, you know, more to the level of perhaps speaking them about business. So talking about AI trust, let's start with that. I think there is a trust gap with AI. Now we published a report on hybrid cloud. We call it the hybrid cloud research report. The 2025 report showed that 92% of organizations had found budget for AI initiatives. Now no surprise there, I guess. But 34% still had major concerns about data security, and particularly privacy when implementing AI. And, I guess we're seeing this play out in real time with customers who want to innovate. They wanna take advantage of AI, but now they need some ironclad governance around their file data and the use of it with AI. And I guess what I would say, Rachel is, I know aid AWS has kinda got extensive AI safety guardrails. That's one of the things, obviously, you get from companies like, Amazon. So you've got services like, Amazon Bedrock guardrails. You've got your responsible AI framework. But what are you really seeing as the biggest trust block is from your side for enterprise adoption? Is it more about the AI models themselves? Is it the underlying, you know, data governance and access controls or something completely different? Security and privacy risks are a big concern from our customers. You know, there's increased pressure to implement safeguards against challenges like model hallucinations, biased outputs, and potential misuse of AI systems. You know, what we're hearing and I'm learning is that inadequate data governance creates additional risks around bias, fairness, and regulatory compliance, while privacy compliance requirements add complexity to data preparation and model development. So what we know now is that without a clear framework for governance and risk management, many organizations, they find themselves hesitant to fully embrace AI's transformative potential. So even as competitive pressures mount to escalate accelerate adoption, we found that, you know, customers are balancing this. Right? Like you said, there's a lot of budget available, but how do they balance this in r and d and making investments with also making sure that the right safeguards are in place? Well, with AWS, you know, you can innovate with confidence. You can take control of your AI with industry leading content safeguards like Amazon bed Bedrock guardrails that give you customizable safeguards and lock up to 88% of harmful multimodal content. Yeah. That's interesting. Because if you go back to our research, only 20% of organizations actually believe their data is AI ready, which is crazy when you think about it. So we're really, I think, dealing with a trust issue that goes deeper than just the AI layer. It's about trusting that the data foundation itself is secure, compliant, properly governed, all the things you just talked about. And the the the need to bridge that kind of gap between data readiness and AI trust. I find that quite interesting. It's something we think about a lot at Nasuni. And and, obviously, it's something that we look to our partners such as Amazon, you know, to to really help allay those fears for the customers. If I move on, to now think about probably the biggest topic for most companies, and that's around ROI and, you know, maximizing the value of their investments. So I think ROI measurement in most companies is the elephant in the room, actually. If you look at global stats, only 2027% of AI projects are actually delivering measurably ROI. You know, measurable ROI is a funny term. It's actually, you know, sometimes with AI, difficult to measure ROI. You know, because when you see in productivity efficiencies, it's very difficult sometimes to get a handle on what the ROI is behind that. But yet, we're still seeing organizations actually increasing their spending to support these AI initiatives. So they're certainly seeing, you know, some payback or expecting some payback for that. But we're seeing, like, when we talk to prospects, they really struggle, you know, not just with really proving the eye value, but with the foundational question of whether the data infrastructure really is ready to support any meaningfully ROI measurement. And I think what it'd be interesting to understand is what metrics you're seeing with AWS customers, using you know, what are they what are they using beyond the obvious cost savings? How are they measuring the fact that actually investing in AI is giving them, you know, better bang for buck? Because it seems like, you know, part of the challenge at a business level is connecting the fact that AI actually is providing, you know, better business payback, and proving that back to the organization. Yeah. No. Thanks, Jim. And I I think of so many different directions that I could go with that that lead up. You know, ROI is, you know, obviously, at the top of every, you know, every business leader's mind. And I think with generative AI AI, it's so, you know, new and developing, and a lot of it started out around national language processing. You know? And they're trying to figure out, you know, that's great, you know, because we can offer maybe customers an interactive chatbot. But, you know, how do we get real value from the data that we have across many years to help make predictions for us from the future that might help us save money in our businesses, may help us plan better? And I think that what we're seeing too is it's very industry specific. So each industry, they're looking at this at, okay, you know, how can we extract value? And, also, how can we prepare the data in a way so that we can train the models and then extract the value from it that we need so then we can start showing ROI? Still feels very new, very early in this this generative AI, evolution, I would say. And and and top of it, the technology is changing so quickly. So companies are also faced with, you know you know, learning new technology, trying to prepare their data, also coming up to speed on the new technology that's coming out just months later. And they're one of the the biggest challenges that business leaders have is they're struggling to identify and prioritize what is the most impactful use cases while also balancing concerns about costs. ROI comes into that and the implementation timelines. So, you know, the challenge becomes more, you know, not just, you know, how do we select the right technology, it's about how do we transform operations, how do we upscale our workforce, and then manage this change across the whole enterprise while maintaining security and compliance. So that's a it's a really big challenge for companies and, you know, it's being required that they transform their business. And to do that, that takes time. And so, you know, we really see that the the the value of ROI is is gonna come across, you know, many years to come. We're just we're just now entering into this phase. We know business leaders are facing unprecedented pressure to adopt and scale AI technologies, but they also have these business challenges internally that they're trying to balance in order to implement them. Yeah. It makes, perfect sense to me. Cost obviously comes into a big part of that. But I think from the Nasuni perspective, just to kinda end that that trend topic, I think what we see is a lot of the ROI struggles often stem, you know, from the fact that they've got data silos and they've got fragmented storage architectures. So we see things like when, you know, when your training data and your production data and your results data are scattered across different systems, it's actually really difficult to measure the impact that I IOI is gonna have because, you know, feeding that back to AI, we've you know, it it involves an integration project. And integration project adds more cost, it adds more time, and actually, it puts a a blocker in the way of really realizing the value. And that's for us at least, you know, that's where we concentrate on trying to, you know, move, companies forward. Are you seeing, you know, the data foundation itself become an inhibitor to, to achieving great ROI results? Yeah. I have a recent customer example where I was talking with a customer who's in the the biomed field. And, you know, their management said we, you know, we need to take advantage of this. We I you know, no matter the cost, tell us what it takes to implement some type of generative AI solution where we can train models on all of our data and help, you know, this this particular customer. They develop new drugs. They take them to market. They wanna reduce the time that it takes these drugs to market. And what they found is that it's not an infrastructure problem. It's like you said, it's a a data problem. You know, as they were interviewing scientists, they under they learned that sometimes scientists capture data on napkins. They don't run the same experiment with the same variable. The word they might capture some data in Excel spreadsheets. There's no standardization. And so they're like, before we even get to, you know, the infrastructure and look at what models and technology, we need help transforming the business on how to capture and the data so that it is, you know, consistent so that we can make, you know, predictions around this and that we can have those those values and insights into our business. So, yes, I see that is, you know, also coming into the industry depends on, you know, maybe, you know, how regulated it is, you know, what type of maybe, experiments they're running very early on versus later in the process, and, you know, the consistency of that. So we we are seeing that with our customers. Yes. Fascinating. I'll I always think it, becomes real when you have a real case study or a real customer in the background. So I think the next trend topic is moving on is we're getting more into the the technical arena now of, like, trends. Mhmm. And, you know, this one is MCP, model context protocol, introduced by Anthropic in September 2024. You know, personally, anybody who follows me on LinkedIn will know that I'm a big believer in MCP being a a game changer because it really standardizes how AI systems can, really access enterprise data sources. Now what's exciting about it for me is, really, for for many companies, how it could solve the data accessibility problem that we see all over the place. So organizations that have issues getting access to data sources and feel that they have to do a, really a connector integration project for every single data source. But they need the data to be able to leverage it, particularly if they're gonna move towards more agent based AI. Now I know AWS has been pushing standardization across some of the services, you know, from Bedrock to SageMaker to the new Amazon queue offerings. How do you see MCP fitting into that ecosystem? Will we see perhaps native MCP support in in Amazon from an integration point of view with with services like Kendra and knowledge bases? Is it something you think, AWS is gonna embrace in in the Bedrock ecosystem? Well, you know, AWS is making quite a few announcements around, you know, MCP and, model context protocol support. In May and in May, AWS announced that we are announced the release of, MCP servers for AWS Lambda, Amazon Elastic Container Services, ECS, Amazon Elastic Kubernetes, EKS, and Finch. MCP servers are, you know, standard interface to enhance AI assisted application development. It equips AI code assistance with real time contextual understanding of AWS serverless and container services, including Lambda, ECS, and EKS. You know, our belief is that with MCP servers, you can get your your you can get from, you know, your idea to your production even faster. You can give your AI assistance access to an up to date framework on how to correctly interact with your AWS service of choice. MCP servers also enable AI code assistance to generate production ready results, by incorporating AWS operational best practices. We have well architected principles and service specific optimizations. When building applications on AWS Lambda, ECS, and EKS, and Finch, developers can use natural language to describe their requirements while AI code assistance handle service configurations. So we really see, you know, that, you know, MCP servers are are gonna help to build and deploy applications and also simplify operations by enabling AI assisted service specific configurations for logging, monitoring, security controls, and troubleshooting failures. Okay. I mean, I I personally think MCP will will help bridge the gap between, you know, enterprise file systems and AI applications. And, I think we'll see customers building custom connectors, pipelines, you know, agent workflows. Because with MCP, we have something that's standardized, obviously has the authentication authorization flow that potentially can help you maintain the data governance between those different steps while still giving, you know, AI the access. Now I would caveat that with, MCP, obviously, is still a fairly immature technology. I think I said earlier that it was only proposed as a standard by Anthropic in September. And we've already seen some fairly high profile, if you like, issues with MCPs that have been released out in the wild by some of the larger companies. And that's not surprising in some ways, you know, not to me at least. I think, you know, the technology there is still kind of bedded getting bedded down. And there are different flavors of how you you can use MCP. If you use it all locally and you use the server locally, you're you're using SDDI O standard input output, you know, a little bit more secure if you trust the m MCP server that you're you're leveraging. You do it all from your laptop. Or, you know, now, Anthropics introduced its, its, you know, streaming m MCP server, you know, released also in the in the desktop in the last few weeks. So I I think, you know, it's gonna be it's it's been huge since it was released MCP. It's kinda spread like wildfire. But I would caution, you know, again, the fact that the the technology is still fairly immature, and we probably will start to see some changes still in that kind of ecosystem. Moving on to retrieval augmented generation. I think for many organizations, you know, RAG or retrieval augmented generation, that's where the rubbers met the road for the last twelve months for enterprises. Pretty much. Even when they don't know they're using rag, they're often using rag because it's kind of been, if you like, you know, democratized. So companies have gone away and built their own rag systems or, you know, leveraged, ecosystem kind of frameworks, you know, like Langchain. But but also they probably leverage things like, you know, Amazon queue, to be able to do that in a much, you know, easy fashion than probably, you know, twelve months ago. But it is kind of exposing also, you know, the weaknesses around unstructured data. And back to that research I talked about earlier, you know, it said that most companies believe only 20% of them think that their data is AI ready. So we've got, you know, a lot of customers out there trying to build right systems, but ultimately on fundamentally flawed data foundations. And and I've worked with, you know, quite a few people and prospects and customers, you know, that it only hits home, you know, the quality of the data once you have it inside of a rack system and you start to do some of those, conversational interactions. Now with things like Amazon bedrock knowledge bases, Amazon Kendra, and now the enhanced rack capabilities in Amazon queue, I think Amazon must see some of that data quality problem firsthand. And my interest here is, you know, how are you helping customers solve that kind of garbage in, garbage out challenge? Is it kind of more about embeddings? You know, is it more technical? Or do you find yourself having to really get involved in the underlying data architecture? Yeah. No. We're always looking for ways to make, you know, to make adoption and to make, it easier for customers to to get value, you know, from the technology for their business. And, you know, this is no different. So what we're finding is that, you know, data transforms good AI into great AI. So, you know, like you said, like, we need to make sure that the data is good in order to have a great AI result. And AWS provides proven, open, and trusted data foundations as well as diverse choices for things like vector databases and rack options across structured and unstructured data and graph data too. So you can really tailor AI to your business. You know, what we find is that operationalizing applications is a continuous challenge, especially for developers who are looking to find ways to integrate generative AI into their workflow. This slows adoption and often requires them changing the way that they work. So they have to work through that as well as make sure that the data is a good quality. As technology advances, you know, developers also have to continuously optimize and maintain their applications with the latest, advancements. So, yes, we're we're looking at ways to make that easier for customers and also while giving them, you know, the broadest choice and, you know, scalability for their solutions. So that leads us on to our last topic. And if, if the preceding twelve months has really been about retrieval of minute generation and how enterprises are looking or have been trying to bed that down, I think agents, or agentic frameworks is gonna be really the big topic for the the next twelve months. And I think agents represent a real shift, you know, from AI that answers questions, you know, the chat bots, to AI that takes actions on enterprise data. And in many ways, I think enterprises are much more comfortable with AI that takes takes actions on enterprise data. And I think that's because they're used to implementing business processes. Business process type technology has been around a long time. The the idea of encapsulating business processes and, automating them has kind of been nirvana, certainly, for my entire technology career in the enterprise. So I think we'll start to see customers going beyond simple rag implementations to agents that actually interact with not just their file system, but, you know, ERP systems, CRM systems. But, obviously, the security implementations of that are pretty big, you know, because, ultimately, you'll have agents that have access, you know, maybe right access to business critical data. Now I'm kind of interested from the Amazon perspective. You've obviously got things like Amazon queue business. You've got your, agent frameworks through Bedrock. They probably I would think starting to get some pretty aggressive adoption. But I'm kinda curious about your approach to agent safety, particularly when they're interacting with real, you know, hardcore enterprise data. How are you thinking about permissions and audit trails and the potential for agents to, you know, maybe make destructive changes to critical data. How is that you know, are there any thoughts around that from the Amazon perspective? Thanks, Jim. This conversation really took us full circle. You know, we started out talking about security, governance, risk mitigation, and ROI. And what we see right now in the market is a significant shift towards AI agents, which is an autonomous system capable of reasoning, planning, you know, taking actions to achieve specific goals across complex multistep workflows. These intelligent agents represent the next evolution beyond traditional single purpose AI tools. It enables sophisticated business process automation and decision making. According to McKinsey research, we found that about 75 of business value generative AI will drive falls into four domains, customer operations, marketing and sales, software engineering, and r and d. So AI agents are particularly transformative in these areas and moving beyond simple code suggestions for developers or ad copy or, yeah, copy generation for marketing teams. What we find is that this evolution is driving new market segments around agent orchestration platforms, multi agent systems, and specialized tools for developing AI systems that can operate independently while maintaining human oversight and control. So, you know, when you're talking about security, security is never a bolt on at AWS. It must be prioritized from day one. AWS is designed to be the most secure cloud infrastructure with built in governance encryption and security controls to help you to help our customers innovate at speed without compromising safety. So, you know, when we look at customers across different industry leading Adgentec applications, we wanna help them get value from from their data now with prebuilt agents. And AWS offers applications with powerful agents available off the shelf that can leverage your data to improve productivity and aid data driven decision making. We also look at developer productivity, and we wanna help developers accelerate building across their organization. And Amazon Q provides powerful agents that accelerate nearly every step of the software developer life cycle and gives developers the time and tools to focus on innovation. We also look at how customers can accelerate and transform with agents. They can save years of development time with agents to accelerate migration and modernization workloads. Amazon applied Amazon Q Agent technologies to Java modernization, saving over 4,500 developers years in analyze comp in analyze compute efficiencies worth hundreds of millions of dollars. So those are some of the ways that we're looking and hearing from customers on how they're able to, you know, get ROI with with agents as well as the security that AWS provides across, our services. Okay. Great. I guess the last comment I'd make on agents is, again, you know, this is a this is something that's evolving and the the this the standards or at least the pseudo standards are still evolving. We talked about model context protocol, you know, which is really the plumbing if you like, the ability to have the you know, what tools can you use and what data or service can you use it against. But, you know, also, there are things like agent to agent, you know, framework, and agent control protocol framework. And this, you know, set of standards that start to become, you know, used and, you know, more referenced, it's certainly gonna be something that I think companies should look at when they're thinking about choosing agent technology. It just leads me to say, thank you very much, Rachel. Really appreciate all your time today. I'm sure that everything you've told us here, particularly about the AWS ecosystem and all of those great topics, you know, people will find, you know, and get amazing value from. So at this point, I'd just like to thank everybody for watching, and thanks again, Rachel. Thanks, Jim. We appreciate your partnership. You know, as always, it's, great to work with you, and thank you for inviting us to participate in the session. And we hope to join you again.