TFiR

Kubernetes (Official) 1.37 lets you scale GPU workloads to zero, so you stop paying for idle pods with no demand. Teams running AI training or batch queue jobs have always needed at least one watcher pod running at all times. That changes in v1.37. HPA scale to zero is now beta and on by default. Hook up external or queue-depth metrics and workloads spin up only when demand arrives. Dipesh Rawat, Kubernetes v1.37 Release Lead at Cloud Native Computing Foundation (CNCF), also covers resilient Watch Cache (now stable) and pod checkpoint and restore (alpha), which lets interrupted AI training jobs resume mid-run instead of restarting from scratch. "This feature is going to save a lot of clusters and GPUs, especially in times where there is a shortage of RAM and memory." In this short clip with Swapnil Bhartiya at TFiR, Dipesh explains what each feature unlocks and how far each is from production readiness. Check out the discussion on our YouTube page: https://lnkd.in/gQinGpPu #Kubernetes #CNCF #GPUCost #KubernetesAutoscaling #AIInfrastructure #CloudNative #PlatformEngineering #DevOps

AI and hybrid cloud are exposing gaps in how IT teams validate failover. Traditional failover testing was not built for the failure modes that come with AI workloads and Kubernetes environments. Teams that do not adapt their validation strategy are accepting risk they may not fully see. Alexus Gore, Customer Experience Software Engineer at SIOS Technology Corp., makes the trade-off clear: "When you're introducing so many more products to create a bulletproof, highly available solution, there's just going to be more room for error." In this short clip with Swapnil Bhartiya at TFiR, Alexus explains why failover validation now demands testing for predictability and adaptability, not just recovery. Check out the discussion on our YouTube page: https://lnkd.in/gytfBCzc #HighAvailability #FailoverTesting #HybridCloud #AIInfrastructure #Kubernetes #DisasterRecovery #SRE #CloudResilience

The new MCP spec eliminates session state — and that changes everything for enterprise AI deployments. Teams building production AI applications have been stuck: the original MCP client-server model couldn't sit behind a load balancer, making scale nearly impossible. The latest specification fixes that by making every request stateless, enabling round-robin load balancing, async task ticketing, interactive confirmation flows, and a one-year deprecation commitment. Du'An Lightfoot, Sr. Developer Advocate - AI Engineering at Akamai Technologies, breaks it all down. "The changes that they have made it more in line with how you would scale an API within your organization." In this short clip with Swapnil Bhartiya at TFiR, Du'An explains what broke in the old MCP spec and what the new one unlocks for enterprise teams. Check out the discussion on our YouTube page: https://lnkd.in/g7zb9EaW #MCP #AIEngineering #EnterpriseAI #AIAgents #DeveloperTools #CloudNative #AIInfrastructure

The AI vs. non-AI application boundary is gone, and most teams have no visibility into it. Legacy ERPs and back-office systems are now inside AI workflows the moment someone points a RAG pipeline at them. Nothing about the application changed. The risk did. Shadow AI is making this harder: developers are using unapproved AI tooling whether policy allows it or not, and platform teams cannot monitor code being generated faster than it can be reviewed. Danielle Cook, Director of Product Marketing at Akamai Technologies, puts it plainly: "You don't have to build AI to become part of this ecosystem." In this short clip with Swapnil Bhartiya at TFiR, Danielle explains why every organization is already in the AI ecosystem and what that means for governance and visibility. Check out the discussion on our YouTube page: https://lnkd.in/gSQGBbkY #ShadowAI #AIEcosystem #PlatformEngineering #CloudSecurity #AIGovernance #DevSecOps #Akamai

75-85% of AI agents run on developer laptops, outside every governance control your enterprise has built. MDM tools manage the device. They do not govern which LLM an agent calls, which MCP servers it connects to, or which credentials it inherits from the developer's terminal. That gap is exactly what Solo.io's Agent Desktop addresses: fleet-level visibility and policy enforcement for agentic workloads running on end-user machines. Michael Levan, Principal AI Architect & Forward Deployed Engineer at solo.io, breaks down the problem and the architecture. "The agent can literally do whatever it wants based on my level of permissions." In this full conversation with Swapnil Bhartiya at TFiR, we explore why desktop agents inherit cloud credentials, where MDM falls short, how Agent Desktop's control plane works, the open source model, and what security teams are demanding right now. Check out the discussion on our YouTube page: https://lnkd.in/gFJaJQtT #AIAgents #EnterpriseAI #AgentSecurity #PlatformEngineering #MCPSecurity #DevSecOps #SoloIO #AISecurity

Every AI agent in production is making model calls all day, and most teams have no idea if those calls are going to the right model. Without intelligent routing, token spend grows fast and stays invisible. The problem is not just cost. It is cost without outcome. Viewers will learn how to measure token ROI, why you should never lock an agent to a single model, and how enterprise governance rules plug into a router. Suresh Mathew, CEO and Founder of Sedai, joined TFiR to dig into all of it. "It's never about just the cost. It's about how much are you spending to get the outcome right." In this full conversation with Swapnil Bhartiya at TFiR, we explore AI model routing, tokenomics in production, hallucination as a design reality, and autonomous vs. human-led model governance. Check out the discussion on our YouTube page: https://lnkd.in/gVGCMxjW #AIInfrastructure #LLMOps #TokenOptimization #AIAgents #EnterpriseAI #AIGovernance #Finops #MLOps

Most enterprise "AI adoption" is just humans forwarding AI output without thinking. This clip cuts through the hype and exposes what real enterprise AI adoption actually looks like in 2026, and why agentic workloads are harder than anyone planned. Practitioners will learn why process debt, not technology, is the real blocker. Rob Hirschfeld, CEO at RackN, calls it the "reverse centaur" problem: organizations using AI instead of thinking, not AI to extend thinking. "Using it correctly and keeping humans in the loop is absolutely essential." In this short clip with Swapnil Bhartiya at TFiR, Rob explains where enterprise AI genuinely stands today and why most organizations are not operationally ready for what comes next. Check out the discussion on our YouTube page: https://lnkd.in/grZB_QHq #EnterpriseAI #AIAdoption #AgenticAI #AIStrategy #CloudNative #DevOps #PlatformEngineering #AIOperations

Agentic AI spends up to 90% of its execution time on CPUs, not GPUs. Most infrastructure teams are optimizing the wrong layer. GPU capacity matters, but tool calling, file writes, and code execution dominate actual runtime in real-world agentic deployments. Here is what that means: end-to-end latency is controlled by CPU orchestration and external connectivity, not token generation speed. Jon Alexander, SVP of Product, Cloud Technology Group at Akamai Technologies, makes the case with data from live customer deployments. "A large portion of the execution duration isn't waiting for the GPU to generate tokens. It's actually in the tool calling." In this short clip with Swapnil Bhartiya at TFiR, Jon explains where agentic latency actually lives and what to do about it. Check out the discussion on our YouTube page: https://lnkd.in/gR5U65yx #AgenticAI #AIInfrastructure #LLM #CloudNative #PlatformEngineering #AIPerformance #MLOps

The patch already exists. Attackers are exploiting the gap before you deploy it. Monthly patch windows were designed for a slower era. AI has changed the math: vendors now discover and fix more vulnerabilities than ever, but malicious actors use the same AI to reverse-engineer those patches within days. Viewers will learn why CVSS scoring is no longer sufficient, how EPSS changes prioritization, and what proactive patching actually looks like in practice. Tuukka Tiainen, Lead Security Engineer at Recast Software, lays out the data and the decisions teams need to make now. "Malicious actors can reverse engineer these patches and even vulnerabilities that seem to have low severity, with the help of AI they can change these exploits. You just need to be patching faster than you used to do." In this full conversation with Swapnil Bhartiya at TFiR, we explore the 2026 CVE spike, patch gap metrics, EPSS-based prioritization, proactive lifecycle patching, and software inventory visibility. Check out the discussion on our YouTube page: https://lnkd.in/gd83HvKR #PatchManagement #VulnerabilityManagement #EndpointSecurity #CyberSecurity #EPSS #AIThreats #InfoSec

The web page is about to become as obsolete as the pre-web Internet. Centralized AI forces users to still click links, paste URLs, and navigate menus. Edge inference eliminates all of that. Dr. Robert Blumofe, Executive Vice President and Chief Technology Officer at Akamai, makes the case that every desktop and web interaction will become agentic, and the transformation will be as profound as the web itself. "The web transformed the Internet and now we've got AI inference transforming the web. I don't think it's that much longer before we're explaining to young people what a web page was." In this short clip with @Swapnil Bhartiya at @TFiR, Dr. Blumofe explains why edge inference enables personalized, conversational AI experiences that centralized architectures cannot match. Check out the discussion on our YouTube page: https://lnkd.in/gsjJVyAB #EdgeAI #AgenticAI #AIInference #EdgeComputing #FutureOfWeb #AIStrategy #EnterpriseAI