Good morning everyone. I'm from Microsoft and I'm thrilled to be here at Comp 42. Network engineering, as many of has traditionally been about cables, routers, and keeping systems online, but today that world is being completely redefined. We're entering an era where AI isn't supporting just the network, it's actively running it. Optimizing performance, predicting failures and adapting on the fly. In this talk, I'll explore how AI is reshaping network engineering, what new skills we need and how we can design scalable, reliable, and intelligent networks for this AI driven future. Let's start with how our field have has evolved. A traditional approach was straightforward. Manually configured devices, monitor up time and react to problems when something broke. If a router failed at 2:00 AM someone got a call. Today, we have shifted to a modern paradigm where intelligent systems can predict issues before they occur, optimize themselves and make decisions based on data. For instance, modern data centers can now automatically reconfigure traffic patterns or, and the paths basically when latency is noticed on a particular path with no human in the loop. So the role of network engineer has changed from being an operator to becoming a systems architect and a data driven strategist. Network engineers must span three worlds, enterprise networks, cloud providers, and telecom operators. Enterprises care most about compliance of time and security. Think of a global bank that must follow strict regulations that keep trading systems live. 24 7. Cloud providers and put emphasize automation and elasticity. Companies like Microsoft Azure or AWS need networks that scale across continents automatically. Telephone operators handle massive traffic and cardiac great reliability. Imagine managing billions of iot connections simultaneously. Each network is unique, yet they are deeply interconnected. The future engineer must pay fuel fluent in all these three. In enterprise environments, the challenge is balancing reliability and regulation. Networks must meet audit data sovereignty and zero trust requirements while staying efficient. For example, a healthcare company integrating patient data across multiple clouds must ensure HIPAA complaints without slowing access to doctors. Engineers have to design architectures that maintain redundancy, security and performance, all while supporting legacy systems that were never meant to talk to modern cloud services. It is like renovating an old house while keeping the lights on. You can't afford downtime in cloud and service provider spaces. Speed and scale is everything. Four imperatives that define this environment first is automation. Tools like Terraform or Ansible replace manual configuration. You can deploy hundreds of routers with a single script orchestration and scale. Managing thousands of network nodes require systems like Kubernetes for networking to maintain consistency and reliability, performance optimization Subsequent. Latency can be made or can be make or break applications like online gaming or stock trading. Elastic scaling networks must expand and contract based on demand. Think of video streaming traffic spiking during a major sports event. AI ties these together, helping us make real time data driven adjustments. Now let's ground ourselves in the core skills that still matter. AI can optimize configurations, but it can't replace deep understanding of writing protocols like BGP or SPF and MBLS, for example, when AI recommends the route change, engineers must interpret, why is it improving latency or masking and underlying topology issue? Likewise, IPV six. Security and observability are becoming fundamental with billions of iot devices. Joining networks IIPV four is simply not enough. So while AI helps automate and enhance our foundational knowledge remains our compass, the network engineer of tomorrow is a hybrid. Part network architect, part software developer, part data analyst. Beyond Protocol Mastery, we need programming, fluency, AI literacy, and data interpretation skills. For instance, when AI models suggests optimizing bandwidth based on telemetry data, you should understand how that model learned, what features it used, and what biases might exist. The more we understand ai, the better we can collaborate with it. Turning the engineer into an AI co-pilot, not a spectator. The toolbox of a motor engineer looks very different. Now we are seeing tools like Ansible, Terraform, net, and Napalm, all focused on automating repetitive tasks. If you add Python, yaml, js O and jet this is now the new command line. Picture this, instead of configuring one switch at a time, you push a GI commit and the change deploy across hundreds of devices, which validation checks built in. That's why the new baseline is clear Python. Get Ansible APIs. SGN controllers, these are the keys to building intelligence scalable systems. AI is already deeply embedded in network operations. It helps with the anomaly detection, spotting unusual spikes in traffic that might signal an attack. Predictive maintenance, identifying ports or links likely to fail soon. Root cause analysis, correlating thousands of logs in seconds to find the issue. Without ai these tasks will become humongous and take a lot of time. But with the help of AI, we are able to scale up much better and root cause it much more efficiently. A real world example is a global Telco used an AI to detect wifi degradation across regions before customers ever noticed. Serving millions in support calls these capabilities transform operations from reactive to proactive, from firefighting to forecasting. The shift AI drinks is not incremental. It's transformational tasks that took hours like sifting through telemetry data now happen continuously and automatically. AI systems analyze traffic behavior, link health and security patterns to adjust configurations in real time. Imagine a network that knows peak hours for your app. Reallocate bandwidth and tunes, quality of service parameters before condition actually happens. That's the power of moving from monitor and react to anticipate and adapt. Self optimizing networks are where AI truly shines. These are systems that can self-diagnose. Radar out traffic around failures, and even heal themselves automatically. For instance, if a link in Singapore fails and the network could re out through Tokyo in milliseconds without anyone touching a console, the impact is massive. Reduce downtime, faster resolution, and lower costs. But autonomy must come with guardrails. Engineers must define safety limits. Like what an AI can change to avoid any unintended consequences. So this definition of guardrails and how you put safety checks in place is pretty important. And this brings us to a cri critical truth. AI can enhance, but not replace human expertise. A lacks context. It doesn't understand business goals, regulations, or organization culture. For example, an AI might flag a large data transfer as security threat, but a human knows it's a sanctioned quarterly backup. The best networks come from AI plus human collaboration. AI accelerates analysis. Humans provide judgment, strategy and ethical oversight. As engineers, we remain the interpreters and architects behind the intelligence feature. Ready networks are modular, API, driven and observable by design. A PA first design allows systems to integrate and automate easily. Microservices architecture ensures scalability. You can upgrade one service without taking the entire system down. Declarative configuration lets you describe the desired state. Instead of typing imperative comments, the system figures out how to get there. Observability by default, gives AI the data it needs to learn and improve. Think of it as moving from a static machine to a living evolving organism. Scalability and resiliency aren't add-ons. They must be baked into the architecture. Horizontal scaling adds more nodes, not bigger ones. Stateless design reduces dependencies. Graceful degradation. If one part fails, the system still runs, even if partially. Chaos engineering deliberately breaking things and testing to build confidence and recovery. Netflix popularized this idea. Their Chaos Monkey tool randomly shuts down systems to test resiliency. Networks need the same philosophy. Plan for failure and design for recovery. So how do we get there? Here's a five step roadmap. Assess current state, identify where AI can help, and what skills your team lacks. Build the foundation, automate repetitive tasks and collect telemetry. Pilot AI solutions. Try AI in low risk areas like anomaly detection. Scale intelligently, which is like bringing successful pilots into production. Monitor results and refine them, iterate over them. Foster innovation, encourage learning hackathons, and crossed in collaboration. This journey is iterative. Each phase strengthens the next. It's evaluation, not revolution. To wrap up, master the fundamentals. Deep protocol knowledge never goes out of style. Embrace continuous learning. AI and automation evolve quickly. Stay curious, think architecturally. Design for adaptability and scale. From day one, collaborate across disciplines. The best results come from networking software and data teams working together. The future of network engineering isn't about replacing people with ai. It's about augmenting our intelligence, creating systems that learn, adapt, and evolve alongside us. Thank you all and I hope this inspires you to start building tomorrow's networks today. Thank you.