Everyone, thank you for joining me at Con 2025. I'm bas, and today we are gonna dive into one of the most exciting frontiers in modern systems, AI driven self infrastructure. Over the next 20 minutes, I'll walk you through how AI is evolving s monitoring and automation into systems that can truly detect, recover, and improve. Just someone who tired of waking up at 3:00 AM incidents in the right place. Let's explore how the future of infrastructure is becoming not only smarter, but truly resilient by design. Let's talk about why this conversation is so critical. Why AI driven self-heal infrastructure is more than just buzzword. We are operating in a world where S isn't optional anymore. It's mission critical. The complexity of what we manage today is staggering. Modern cloud infrastructure spans thousands of interconnected services. Manual oversight just isn't scalable anymore. Even the best engineers can't babysit every node, every container, every function, 24 by seven. On top of that, our current systems are still largely reactive. We wait for alerts, we jump on incidents, we write postmortems, and in high availability environments. Especially in healthcare, finance or e-commerce, that model just doesn't cut it. Downtime isn't just inconvenient. It's expensive. We are talking about millions in lost revenue annually, or even brief outages. That's why the shift matters. AI changes the paradigm. Instead of responding after things go wrong, we can predict, prevent, and heal before users are even impact. This is where we even move from firefighting to future proofing. Today, I want to show you how exactly we got it. Let's take a quick look at how SRE has evolved over the years, because understanding this evolution gives us context for where we are headed. We have started with SRE 1.0, a phase with operations. Were largely man engineers. Were constantly on call. Basic monitoring tools helped, but the response models are still reactive. This is also when we first saw the idea of error budgets, helping balance reliability with innovation. Then came SRE 2.0 where things really started to shift. We embraced automation with infrastructure as code scripted remediation and alerting. That got a bit smarter. Chaos engineering also started to emerge. Testing resiliency proactively. We moved from just monitoring to observability, not just knowing something broke, but wide broke. Now we are entering SR 3.0, a new era where systems are becoming autonomous. We are talking about predictive healing, where AI anticipates problems before they happen. Engineers don't just respond anymore. They architect systems that respond themselves. It's a strategic role now, designing guardrails, training AI models, and ensuring systems evolve safely. And this shift into AI native SRE isn't just exciting. It's necessary because systems are scaling faster than humans can keep up. So let's explore what makes AI powered self-healing possible. Now, let's. It's not magic. It's a structured progression that builds intelligence into the systems stage by stage. It starts with detection. AI systems, monitor logs, metrics and signals, identifying anomalies that a human operator might never notice. We are talking about subtle changes in behavior where edge failures or slowly building patterns that don't. That feels slow, but never breaks during the demo. AI catches those moments. Next, machine learning kicks in with production models are trained on historical data and live telemetry. The forecast failures are sometimes even days before they actually happen. This lets us shift from reactive firefighting to preventative action, then comes remediation. This is where AI really flexes. It can autonomously trigger corrective workflows. Whether it's restarting services, draining traffic from a failure node, or scaling a specific region, the goal is rapid resolution with zero human intervention. And finally, optimization. Each incidents become a learning opportunity. AI systems use postmodern data to refine future predictions, reduce noise, and adapt to all infrastructure behaviors. This feedback loop creates a smarter, more resilient system over time. All four stages, detection, prediction, remediation, or optimization build on each other. In mature systems, we are seeing up to 85% incident healing accuracy without human involvement. That's the real magic, continuous improvement without burnout. Let's take a quick look at how one of the biggest streaming platforms in the world. Netflix is putting AI powered self-healing into action. Netflix uses a tool called Chat the Chaos Automation platform. It leverages machine learning to proactively search for resilience gaps before real users even notice them. Think of it like a in very intelligent troublemaker in the system. It tries to break things on purpose, but only to make even everything stronger. It sharper the false it injects Aren random. They're based on calculated risk models, simulating the most likely points of failure with surgical precision. It's not chaos for chaos sake. It's chaos with intent. What about the results? The speak volumes. Netflix cut their entity by 30 to 50% and proactively prevented over 200 outages in just one year. In just 2023. That's hundreds of moments where you didn't even get the spinning buffer icon mid episode, and that's a win for both reliability and user satisfaction. And one of the most powerful outcomes. SRE now focus less on firefighting and moral architecture. They are not just reacting, they are building proactively with strategy. And now let's shift to another tech giant. Where the focus is on precision at scale, meta runs tens of thousands of servers and detecting subtle behavior failure signals across all of them is like spotting a signal, single flickering light bulb across an entire city. That's where machine learning steps in analyzing telemetry, where millions of components catch signatures before they escalate. Here is where it gets futuristic when the AI reaches 85% confidence. A failure is imminent. The system doesn't wait. It proactively migrates or migrates workloads long before any human would typically react to it. It's like watching the weather and evacuating before the storm instead of rebuilding after it hits the outcome. 76% of potential failures preventive. A 23% boost in capacity planning, and over $45 million saved annually. That's not just resilience, that's remarkable. ROI. And that little chat you see there, that's the AI predictions, outpacing actual failures. It's not perfect, but it gets close. And at meta scale even, a few percentage points matter. Let's shift our focus to Microsoft Azure, a company operating at massive global scale managing services for millions across industries. Fully AI driven self infrastructure and the results are both measurable and inspiring. First, they have seen a 65% reduction. That means AI is filtering of the A charter and surfacing only actionable incidents. Think of it as noise canceling headphones, but for your operations team. Then comes uptime improvement, jumping by 35%. Autonomous remediation means systems are no longer waiting on a human to jump in. They're healing themselves, keeping services running smoothly. The most impressive number, 90% auto resolution. That means the vast majority of common incidents are resolved entirely without human intervention. For sre, this translates to a shift from meaningful. Of course, cost reduction. A leaner smart infrastructure leads to a 44% drop in operational expenses. That's the kind of number that speaks to both the tech teams and as well as the CFO. This case highlights what's possible when AI is embedded, not just as an assistant, but as an autonomous partner in infrastructure resilience. Azure, just automating task. It's reimagining the. Now we have seen how AI can transform infrastructure, reliability, and resilience. But let's ground ourselves for a moment. The reality is AI in SRE isn't all magic and automation. It comes with a unique set of challenges that we must be intentional about first. Explainability AI systems often behave like black boxes. They make decisions and the team is let wondering why. This creates accountability gaps, especially during incidents when every second and every insight matters. Second, human oversight. Finding a balance between trust and control is tough too little automation, and we miss opportunities for efficiency too much, and we risk over resilience. Something that can backfire during edge cases or. AI systems, infrastructure level across access becomes new attack surfaces. We're not talking about ML poisoning, where the model itself becomes the vulnerability. This isn't science fiction anymore. It's an emerging reality. It finally, bias and blind spots, your model is only as good as the data it learns from. And if your data carries historical biases or lacks representation for each cases. The system will inherit those flaws. This can lead to unequal reliability outcomes. So while the potential of AI and SI is massive, it depends. It demands maturity, caution, and clarity. If we don't solve these challenges with intentionally, we are simply trading one set of problems for another. Take moment. Zoom this going. Assist mostly with anomaly detection and noise reduction. It's helpful, but still reactive. In the next two to three years, we enter the transformation phase. This is where AI begin handling known issues autonomously. Auto healing becomes a norm for common failure patterns. It's not just alerting anymore. It's fixing. This shift will begin to reshape how S teams are structured. From reactive firefighting to strategic oversight roles, as we mature into phase three, AI native reliability platforms will dominate. This means SRE won't just manage infrastructure. They will take on new roles like training AI systems, evaluating critical decisions, and setting governance policies. The reliability stack will be AI first by design, and finally, we'll reach the reinvention phase. Where AI systems don't just follow rules, they evolve, they learn, adapt, and redesign their own operational processes. Human engineers will take on higher roles guiding innovation, defining ethical boundaries, and managing the complex relationship between autonomous systems and human oversight. If we think, if it think of it as a timeline. Now AI assists next three to two. Two to three years. AI auto heals. Five to AI infrastructure that learns, and this isn't Afic science fiction, it's organizations, and let's bring AI. It transforms, SRE from reactive support to a forward looking discipline that proactively ensures reliability at scale, SREs become architects of resilience. Second, human element. AI doesn't replace us. It needs us. Engineers remain essential to tune models, validate decisions, and ensure outcomes aligned with business and ethical expectations. Move from scripted writers. Signal interpreters from hands on troubleshooters to strategic AI trainers. Third, start small. You don't need a moonshot. Begin with your biggest pain point. Observability, remediation, or ity. Even incremental AI integration here can yield massive ROI As and as trust goes. So does the, and this part is important. AI doesn't. It's a force multiplier, not a substitute. What matters most is thoughtful implementation, not flashy use cases, but focused impact. That wraps up our journey into AI driven self-healing infrastructure. Thank you so much for your time, your attention, and your curiosity throughout this keynote. I hope this gave you fresh insight into where site reliability engineering is headed. Ideas, challenges, conversation, going, thank.