AI As The New Enterprise Utility | Agentic AI Podcast by lowtouch.ai

We've got a fascinating deep dive today. It's all about a concept that really redefines how organizations, especially large enterprises, are thinking about artificial intelligence. For years, AI was kind of stuck in the land of standalone pilots, right? Just these bespoke departmental projects. - That's absolutely right. - But the sources we're looking at today argue that AI is rapidly becoming something much, much bigger, a foundational, almost invisible utility layer. It is. It's shifting from an application you decide to use to essential infrastructure you just wire into your operations. - Okay, so like the shift from having your own generator to just plugging into the grid. - Exactly. That's the perfect parallel. Or moving from old point to point network connections to just having always on broadband. AI is moving to that invisible utility layer that just quietly powers how all the work gets done. - So for you listening to this, whether you're a CEO or a team lead, the question isn't whether to use AI anymore. It's how to wire it in. Safely, reliably, and at scale. - That's the challenge now. - Okay, let's unpack this then. AI as a utility. What really defines that? What makes it a utility instead of just another project? - The sources are pretty clear on this. They define the path along three critical dimensions. If a technology doesn't hit all three, it's still a project, not infrastructure. - Give it to the framework. - A true utility has to be first, reliable, second, metered, and third, ubiquitous. And AI is now finally evolving to meet all those criteria. So first up, always on. Modern enterprise AI platforms, they run models continuously. They're behind core functions like search, pricing, logistics, customer support. We've moved way beyond those old ad hoc pilots. It has to be as dependable as your OS. - And that always on piece that implies a massive infrastructure commitment. - It does, and that brings us to the second point. It must be metered and scalable. This is really where the economics start to make sense. Things like model calls, vector queries, GPU time, they're all becoming metered resources. It's just like how IT has always tracked CPU cycles and storage. - Hold on, let's quickly define that jargon, a vector query. We all kind of get CPU cycles, but what is that, and why is it a metered cost now? - That is a great question, because it really defines the new cost structure of intelligence itself. A vector query is how you search for meaning, not just keywords. So instead of searching for an exact phrase in a database, you're searching for the closeness of concepts in all your unstructured data. And the ability to do that billions of times for personalization, for document retrieval, that consumes a ton of specialized GPU power. So it has to be metered. - So the only reason high volume use is even possible is because the unit costs for that computer dropping, just like they did for storage. - Exactly, and that leads to the third dimension. It has to be abstracted from end users. - Your employees don't go to an AI tool anymore. They just see better outcomes, faster answers, fewer errors, better recommendations, all inside the tools they already use, all the complex work, the orchestration, the governance, it lives in the platform layer, invisible. - That extraction point feels crucial. It means the metric for success isn't logins to some new tool, but just better business outcomes everywhere. - Precisely. - But what is the immediate financial driver here? I mean, this is a huge infrastructure overhaul. It requires real, C-suite buy-in. - And that raises the key question. What is the ROI that's driving this now? Not five years from now. And the sources are very specific here. The data shows the economic impact is already profound, massive cost savings, and new revenue generation. - Let's start with the cost savings set. That's always the easiest sell. I imagine this is really hitting in big, complex industries, like finance. - Banking and financial services is the textbook example. The sources we have set projections that as AI scales across the whole stack, it could trim industry costs by up to 20%. - 20% sector that large is enormous. It lives good, deeper. You mentioned multi-agent AI systems. How are they getting that staggering 20% to 60% productivity gain for, say, a credit analyst? What's the actual mechanism? - It's the difference between simple automation and coordinated intelligence. A multi-agent system doesn't just fill out a form. It handles a whole complex workflow. - So it's more than just RPA? - Far beyond it. For that credit analyst, the AI agents are collecting data from all these different sources, their cross-checking regulations, generating the first risk summary, and then they route the whole package to the human for the final strategic sign-off. - So the human is freed from all that data gathering? - They're freed from the 80% of their time spent on data, assembly and verification, and the result is 30% faster decision-making and critically better risk insight. - That completely explains the productivity leap. It just shifts the analyst's role entirely. And we're seeing this in software development too, right? Where one bank saw a 40% productivity boost. - Exactly. Now look at insurance, claims processing. It's a massive high volume, rules-based workflow. - A perfect utility use case. - Perfect. AI acts as an always-on-layer reviewing filings. - Gartner-cited figures show AI can slash claims processing times by around 30%. And cut the cost of that processing by up to 40%. - That's significant. And I assume the same utility model applies to things like fraud detection. - Oh, absolutely. Fraud analytics is a classic utility. It's just continuously monitoring data streams that can cut claims handling costs by another 30% by catching fraud before a pay-out ever happens. - What stands out to you about the impact on health care? That's a sector with, I mean, just massive spending challenges. Potential savings are just enormous. Some studies estimate wider AI adoption could save roughly 5% to 10% of US healthcare spending. - Which is what in real dollars? - We're talking 200 to 360 billion dollars annually. By automating admin tasks, optimizing hospital resources, and providing clinical decision support. - And the sources connect that economic benefit directly to better care, which is so important. - Yes, and this is critical. Use case analyses show AI could reduce medical errors by over 80% in some scenarios. So when you get massive cost reduction and a huge improvement in patient safety, AI as a utility becomes a societal imperative. - Right, not just a corporate efficiency play. And we see this in other infrastructure sectors too, like energy. - Absolutely. For great optimization, smart grids, one major international initiative estimates AI could unlock up to 300 billion dollars in global efficiency gains this decade. - Oh. - Through automated load balancing, better demand forecasting, again, a classic meter, always on utility that's coordinating everything. - Okay, let's pivot. Let's move away from cost cutting and talk about revenue. Where does this utility layer translate directly into more sales? Retail-need commerce seems obvious. - Correct. The personalization layer itself is now utility. Companies that use AI-driven personalization, they often see a 10 to 20% revenue lift. - 10 to 20%. - And some report about 20% higher sales from that alone. - It can boost profits by about 15%, and even marketing costs by roughly 20%. - So if personalization is now this ambient utility that every retailer can plug into, how do companies differentiate if everyone has the same intelligence layer? - The differentiation moves. It's not about having the tech anymore. It's about how quickly and deeply you can integrate it. The sources show that AI recommendations already drive more than a third of some e-commerce revenue. The winners are the ones who can integrate that utility seamlessly across every single touch point. - And it's the same story in media and entertainment, I assume. - Even more so. Strong personalization leaders in media, they generate around 40% more revenue from those capabilities than their peers who are lagging. The recommendation engine is the utility that dictates your subscription revenue. If it's down or slow, you lose subscribers instantly. - So if we pull all this back to the big picture, we're seeing 20 to 60% productivity gains, 20 to 40% cost cuts, 10 to 20% revenue lives. What does that range of numbers tell us about where AI is today? - It tells us that we've crossed the threshold. The unit cost of AI compute is falling so fast and the efficiency gains are so proven that the shift at AI as a utility, well, it's not optional anymore, it's mandatory. - And to achieve those kinds of economic numbers, you can't just bolt on a pilot project. The sources are pretty clear that enterprises need a whole new technical foundation. - So what are the key shifts that define this new AI utility stack? - The scale we're talking about, it mandates a switch from departmental tools to true infrastructure. And that transition is defined by six, six critical technical shifts. First is the move to private AI infrastructure. - Private, so not just public APIs. - Right, companies are building platforms that combine on-prem, private cloud, and VPC hosted models to keep control. - Why is that VPC, that virtual private cloud, so important here, why not just use the big public cloud APIs for everything? - Because a high volume utility needs two things, low latency and strict data control. A VPC gives you this security and the predictable performance you need when you're running millions of confidential queries an hour, especially in regulated industries like banking or healthcare. - Got it, control. Okay, what's the second structural shift? - Second is a gentick automation. This is where the work gets done. This is what delivers those 20 to 60% gains. We're seeing multi-agent systems emerge that can coordinate complex end-to-end workflows. - So not just one step, but a whole sequence of tasks. - Exactly. It makes decisions, it uses multiple models. It's far beyond what traditional RPA can do, which just follows fixed rules. Adjunctick systems can reason. - Okay, so that's a big leap. What's third? - Third is the move to multi-model ecosystems. Enterprises are mixing and matching, large foundation models for general reasoning, domain-specific models for expertise and small, efficient models for simple tasks. The utility layer just intelligently routes the workload to the right tool for the job. And speed is everything for utility. If it takes years to deploy, it's useless. Which brings us to number four, rapid deployment cycles. Modern AI platforms are aiming for weeks, not years, from an idea to actual impact. They do it with reusable components and APIs so you can continuously improve the service. And what's enabling that faster adoption? Number five is no code orchestration. Your business teams, your ops teams, they can now configure and orchestrate these complex AI workflows without having to write custom code. It's like what low code did for process automation. It just democratizes access. And finally, you have to be able to trust it. That's the sixth and final shift, secure govern services. Centralization is absolutely key here. Policy observability, guardrails. It's all centralized to ensure AI is used in a compliant, auditable way. Which means companies have to seriously invest in reliability, SRE, and AI ops. Now we see those acronyms a lot. In the context of an AI utility, what do SRE and AI ops actually mean? SRE, site reliability engineering means that when the AI utility fails, you don't just fix it, you automate the fix so it never happens again. AI ops takes that a step further. It means you use AI to manage the AI utility itself. So AI managing AI. Particularly predicting outages, diagnosing performance issues, optimizing resources, automatically. You simply can't run a foundational utility at the scale without that kind of robust automated reliability. This really sounds like a move away from those huge multi-year transformation programs. It's more about continuous incremental deployment, plugging capabilities into existing systems. Precisely. And the utility models, what makes that sustainable. You look at the government and public sector early models suggest these gains could reduce budget deficits by hundreds of billions over a decade just through more efficient administration. But that only happens if the AI is treated as a secure always on utility. So given that this shift seems inevitable, how should enterprise leaders be preparing for this right now? They have to stop treating AI as a bespoke technology and start designing their environment around this utility model. The force is 0.5 foundational steps. First you have to define the AI utility layer, set clear boundaries for this shared, metered resource. And then focus the investment where you know you'll get a return. Correct. Step two. Prioritize high ROI workflows. Be ruthless. Focus on the areas where the sources show those 20 to 60% productivity gains are most likely. Then step three, invest in reliability, SRE and AI apps like we just discussed. Make reliability a core metric from day one. What about the biggest roadblock for most companies? Old data, legacy systems. That's step four, modernize data and legacy systems pragmatically. The goal isn't a total rip and replace, it's connectivity. You have to make sure the utility layer can access the data and needs without demanding a disruptive overhaul of everything else. And then finally step five is to build governance and trust from day one. Centralize everything. Centralize policy, guard rails, transparency, it has to be compliant and auditable from the very beginning. You know, this is where it gets really interesting for me. This isn't just about using a new feature, it's about redesigning the entire company around ambient intelligence. The idea that scalable intelligence is just there, powering every decision. Invisible but essential. Exactly, the sources are so clear on this. Leaders need to stop chasing isolated pilots and start designing an enterprise AI utility layer now. That's how you capture the immediate cost savings while building the foundation for the next decade. So this deep dive really shows the question is settled. AI is becoming foundational infrastructure. And the sources show us how this utility metered reliable abstracted is what delivers those incredible productivity gains of up to 60% and revenue lifts up to 20%. What's fascinating to me is that the requirements those six technical shifts are now so clearly defined. It's not a technological mystery anymore. Companies know they need private infrastructure and agent to automation. The problem is shifted to one of architecture and change management. So what does this all mean for you as you think about your company's strategy? The sources really emphasize that success won't come from just acquiring models. It will come from building that secure, governed infrastructure immediately. Treating AI as a shared service, not a departmental experiment. Which brings us to our final provocative thought for you to mull over. If the cost of AI computation keeps falling like storage and networking did, and if AI agents can automate 20 to 60% of labor workflows. Which business function, one that we all currently consider essential, will be the first to entirely dissolve into the invisible utility layer of the enterprise.