In episode 65 of The AI Forecast, "The Vibecoding Liability: How Unchecked AI Can Kill Cloud ROI," David Linthicum joins host Paul Muller to reveal the hidden costs of hybrid and multi-cloud environments and explain why cloud governance and resilience have become boardroom priorities.
As high-profile cloud outages expose hidden dependencies and single points of failure, IT leaders must rethink resilience, data management, and accountability across hybrid cloud environments.
Here’s what stood out from Paul and David’s conversation:
The Core Differentiator: Reliability Versus Resilience
Paul: Resilience is a funny word because I think people often equate resilience with reliability, and there's a really big difference, isn't there?
David: There is. I mean, resilience is your ability to not let these disasters stop your processing and your business. In other words, what are plan A, plan B, and plan C? How resilient and fault-tolerant is this going to be? Reliability is basically about a component: how well it's going to maintain itself, and if it's going to fall out of place, how to recover from that. Resilience is your responsibility, reliability is not. Typically, if it's with a cloud provider, it's their responsibility, but you'll still be affected. You're going to be paying the bill. There's no credit you're going to get from these cloud providers when they go down.
Paul: Resilience is an architectural artifact, not a consequence of a component, isn't it? It's how you design your system. It goes back to that enterprise architecture.
David: It's all architecture, and it's on the application and enterprise layers. You've got to build and plan for resiliency. It won't happen automatically, and it’s not contained in the clouds. That's where people were surprised. They thought they would be completely resilient to any issues they have, but now they realize they're fallible like everybody else. Part of building an AI system, an enterprise architecture, or any kind of architectural planning is about resilience. It's as important, if not more important, than security, governance, and the other things we have to go through. It has to be operationalized so you can actually prove, with metrics, that this thing won't stop the business from processing if the worst happens. And you basically have to spend the money and time to figure that out.
If you don't have resilience, you won't be able to recover from these kinds of things.
Accountability and Observability in a Hybrid World
Paul: Now, a lot of people are talking about hybrid clouds, but it seems, in some respects, to be a combination of the best and worst attributes of both on-prem and cloud worlds. How do we build clear accountability and observability in what will ultimately be a hybrid world?
David: If you're building hybrid and multi-cloud solutions, you have to basically manage the complexity that's part of the solutions, and resilience is going to be a common control plane that goes through that. People think, "Well, I'm going to build this thing in a hybrid way where I'm going to be able to fail over to my on-prem systems, or even fail over to another cloud.” That's perfectly fine, and it works, but it's going to cost you money. I think the ability to understand what those costs and resources are, and how to manage them, becomes the biggest point of contention.
Multi-cloud is great because you're allowed to use the best technology to build more efficient systems, but resilience and reliability are going to be issues within those architectures. I always say, you can have resilience, and you can have efficiency, but you can't have both. We either have to build the architecture for resilience, or we're going to have to deal with outages three or four times a year that will cost the business billions.
Rising Cloud Costs and The Repatriation Trend
Paul: With regards to things like catastrophic outages, cost overruns, and complex accountabilities, it's not surprising that a lot of companies are thinking about repatriating workloads. What's the state of play there, and what are some of the struggles people have as they're trying to bring some of those workloads back on-prem?
David: The big thing would be the cost of doing it. There are two layers there. Number one, you’ve already spent around half a million dollars on applications and migrating everything to the cloud, and now you’d need to spend a similar amount to move it back. Second, you’d have to go to the board of directors and explain that decision and the path forward. That’s a difficult conversation, because it means acknowledging that the move to the cloud, which was originally expected to be more valuable and reliable, didn’t deliver as planned. Someone will have to go hat in hand and explain that, as a result, the organization needs to shift back to an environment where it has more control over the hardware.
Typically, going to colocation providers and managed service providers is much more efficient, but they're reeling from the cost of the cloud. And now that they're looking at the AI workloads, they're trying to make that move even quicker because they can't afford the cloud. Even though the cloud's going to be the easy button for AI, it's the path of least resistance for building these systems. You get a whole ecosystem ready to go on demand, but it's too expensive for most enterprises. If we're going back there for economic reasons, then we have to put some resources in place to ensure we do it effectively.
Paul: How many other developers in how many enterprises have spun up a little side project in a vibe coding app that's generated, you know, incredible compute workloads or storage workloads that are resulting in cost overruns?
David: You're coding by telling the AI system in terms of what your interpretation is and what they need to code. And the thing is it doesn't understand the nuances there. It doesn't understand how to deal with the efficiencies and you end up spending more money. And so that kind of stuff, the vibe coding stuff, you know, it's fun to think about, but the thing is we have to get some human control over these things. And the more I see these coding systems that go out and you know most of my clients are trying them, they're failing because they're not able to get to the efficiency that they need.
Catch the full conversation with David Linthicum on The AI Forecast on Spotify, Apple Podcasts, and YouTube.
