Abstract

Every modern enterprise is rushing to capture AI ROI, yet the vast majority of deployments stall out due to a silent internal crisis: accelerating Shadow AI, fragmented data governance, and intense security anxieties. SREs and platform architects are routinely caught in the crossfire—tasked with scaling intelligent systems while defending highly complex, distributed data boundaries. The root cause isn't the technology; it’s a structural vacuum. Most organizations are completely missing a centralized AI Center of Excellence (CoE) to steer the ship. This talk delivers a highly practical executive framework built directly on top of Google Cloud Professional Services Organization (PSO) battle-proven practices. Moving far beyond the AI hype, the session will unpack how to balance Technology, People, and Processes to transition your enterprise from fragmented, high-risk operational pockets into a secure, transformational powerhouse.

Bio

Roman Sorocan is the Global Head of Google Cloud at DataArt, where he leads the global scaling of cloud practices across the finance, healthcare, retail, and travel sectors. A dedicated futurist and strategic consultant, Roman empowers enterprises to navigate disruptive technological change through applied AI, robust cloud solutions, and future-of-work strategies. His career includes over six years at Google, as well as serving as Director of Google Cloud at Searce. Roman is also an active Angel Investor focused on autonomous AI infrastructure layers. Aligned with Google Cloud’s battle-proven adoption frameworks , Roman combines deep enterprise technical architecture with a core "People First" philosophy to help organizations deliver AI with real, sustainable ROI.

Abstract

The GenAI observability space is fragmented right now. OpenInference, OpenLLMetry, framework-specific conventions are all solving the same problems with incompatible attribute names. That made sense when OTel’s GenAI support was thin. It makes less sense today. OTel is where this converges. Getting there from where most teams actually are isn’t obvious. Kasper and Adriana cover the current landscape, how the genainormalizer processor bridges the gap at the collector layer, and what a realistic path to OTel-native GenAI observability looks like. Then the harder question: your agent just deleted a database. What does your telemetry actually tell you? Non-deterministic systems don’t leave obvious footprints, and most teams discover that at the worst possible time.

Bio

Adriana Villela is a Principal Developer Advocate at Dynatrace, where she focuses on cloud native technologies, observability, and helping developers build and operate reliable systems at scale. She is a CNCF Ambassador and a maintainer for the OpenTelemetry End User SIG, actively contributing to the open source ecosystem and community education. In addition to her advocacy work, Adriana hosts the Geeking Out Podcast, where she explores technical topics and career stories with practitioners across the industry. She also writes regularly on her personal Medium blog, sharing practical insights drawn from hands-on experience. Based in Toronto, she works remotely and remains deeply engaged with global developer communities. Kasper Borg Nissen is Director of Product Marketing & Developer Relations at Dash0, CNCF Ambassador, and Co-Chair of KubeCon + CloudNativeCon Europe and North America. A long-time cloud native community leader, he co-founded Cloud Native Nordics, organizes Kubernetes Community Days Denmark, and is a frequent international speaker on observability, platform engineering, open source, and cloud native technologies.

Abstract

Observability is undergoing its biggest transformation since the invention of distributed tracing. We're moving from passive dashboards that humans must interpret toward autonomous AI agents that diagnose, recommend, and eventually remediate. This talk traces the architectural patterns and ML foundations enabling this evolution. Drawing from three years of building production AI observability systems, including an 18-model consensus engine achieving 10⁻³⁶ theoretical false positive rate and a composable agent framework powering 22 specialized troubleshooting agents, we'll explore: Why consensus-based ML beats threshold alerting: The mathematical case for 18-model ensembles in production Multi-agent orchestration patterns: When to use advisors, routers, and handoffs for complex troubleshooting The road to AI Co-SRE: What's achievable today versus what's coming in 2026-2027 Practical implementation: Avoiding the pitfalls that derail most AI observability projects You'll leave with a framework for evaluating AI observability tools and a realistic assessment of where the technology stands today.

Bio

Shyam Sreevalsan is a Product, Strategy, and AI leader currently leading Product and AI initiatives at Netdata. His focus spans real-time observability infrastructure, ML-assisted anomaly detection, and edge computing for autonomous robotics fleets. An alumnus of BITS Pilani, Shyam is also the creator and writer of The Day After Tomorrow, a biweekly newsletter exploring future technology trends and AI transformation.

Abstract

At scale, human review of Terraform plans stops being a safety mechanism and starts being a bottleneck. Today's instinct is to delegate to AI, but AI review breaks compliance requirements, can miss edge cases, and removes human accountability from a pipeline where accountability matters. This talk is about a third path: evaluating Terraform plans programatically and deterministically using conftest. Policy-as-code that you can reason about, version, test, and trust, not because a model said so, but because the rules are explicit and the outcome is reproducible. As AI-generated infrastructure changes become routine, safe auto-apply isn't a nice-to-have, it's the next thing you need to build into your infrastructure pipeline.

Bio

Ricard is a Lead Site Reliability Engineer at Cisco ThousandEyes' SRE team. He is responsible for a Terraform pipeline with 500+ developers, over 140k resources under management, and multiple thousands of plans a day. Ricard is currently writing a book about homelabbing, so go talk to him if you have a homelab!

Abstract

Ephemeral Environments (EEs) can dramatically improve reliability: they help prevent issues from reaching production, reduce shared environment failures, and enable safer debugging when incidents happen. But what happens when we want to incorporate this strategy on complex architectures? At that point, interesting challenges start to appear: stateful systems and databases, long environment startup times, multiple dependent services, different use cases (PR validation vs release-level testing), and so on. In this talk, we’ll explore how EEs behave outside the happy path. We’ll look at real architectural trade-offs and patterns, including hybrid approaches where ephemeral services interact with controlled data layers. Rather than presenting a perfect solution, this session focuses on the decisions and compromises required to adopt EEs beyond simple use cases.

Bio

Marcos Novelli Harispe is a Fullstack Engineer and ORT University Lecturer focused on scalable AWS cloud systems. He specializes in event-driven architectures, Data Lakes, and infrastructure as code (Terraform), with extensive experience migrating production microservices to serverless. Equal parts hands-on engineer and educator, Marcos is passionate about reliability, observability, and mentoring the next generation of tech talent.

Abstract

Agentic AI and AI SRE is everywhere in demos; but production-grade, self-healing operations are rare, because an agent can only remediate what your systems let it understand. This talk argues that reliable self-healing starts at design time, with AI-native engineering: build applications that emit the right signals (meaningful SLIs, structured events, rich context) so AI agents can detect, diagnose, and safely remediate issues on their own. Drawing on real-world experience delivering production systems in complex, regulated environments, we’ll cover: How to instrument for autonomy rather than dashboards, How to keep agents inside guardrails with auditability and human oversight built in from day one How self-healing translates into measurable outcomes: less toil, lower operational cost, and higher reliability. Attendees will leave with a practical approach to designing signal-rich applications and introducing safe, production-grade AI-driven self-healing into their own platforms.

Bio

Luca Lanziani is Head of DevOps and Platform Engineering at Nearform, where he helps engineering teams use AI in development in ways that actually improve delivery; not just how fast code gets written. His work sits at the intersection of AI Engineering, Platform Engineering, and software delivery, with a focus on the full path from intent to production: specs, agent workflows, review, testing, deployment, and clear ownership once software is live. Luca builds his AI Engineering work on a deep platform and DevOps background: workflows, review systems, automation, and accountability, so AI-assisted development holds up in real organizations, not just demos. Deepak is the Global Head of Consulting at Nearform. He’s a technology transformation leader who specialises in delivering large-scale platforms and solutions across data, AI, cloud and modern engineering practices. He’s helped deliver complex change at enterprise scale for some of the world's largest brands, lately with key focus on AI Native Engineering and AI SRE.

Abstract

Just run the pipeline again! An expression not unfamiliar, but have you ever considered how wasteful it is? With GreenOps you can reduce the environmental impact of your pipelines so that re-runs are less costly! In this talk, I’ll demonstrate how to build a CI/CD pipeline that uses energy-emission data to decide when it’s best to run. You’ll learn how to cut down on compute cycles by running only the tests actually impacted by a change. I’ll also show you how scanning your deployment templates can reveal more sustainable options to improve your deployments. The time to worry about our environment is here already! So why not make your pipelines greener, reduce your carbon emissions, and save money at the same time? Join my talk and find out how!

Bio

Active in the IT industry since 2012, Leo Visser is a Subject Matter Expert for Azure and AI Foundry at OGD. He advises organizations on AI, automation, and cloud architecture, combining tools like PowerShell, Power Platform, Azure Logic Apps, and native Azure services. His work strongly focuses on security, sustainability, and long-term value. Leo is a Microsoft PowerShell MVP.

Abstract

Your Internal Developer Platform lives in two worlds. In the Right Side Up, developers rely on self‑service observability and resiliency to build, deploy, operate and debug their apps. In the Upside Down, the platform itself must be observable to ensure its reliability, performance, resiliency and to understand usage patterns and adoption. The twist? these two worlds are not separate. Improving observability and resiliency of the platform directly improves the observability and resiliency for the applications built on top of it. Just like in Stranger Things, observability and resiliency becomes the bridge that connects both realities In this talk you learn real-world OpenTelemetry-based best practices to observe the core layers of your platform such as your Git, Argo, Backstage and k8s! You learn how to derive SLOs from that data to increase resiliency for your platform and the apps and services deployed through your platform!

Bio

Andreas Grabner (@grabnerandi) has 25+ years of experience as a software developer, tester and architect and is an advocate for high-performing cloud scale applications. He is a CNCF ambassador, contributor to the CNCF project keptn, Co-Author of Platform Engineering for Architects and a Fellow DevRel for Dynatrace. Andreas is also a regular contributor to the DevOps, Platform Engineering and Cloud Native community and a frequent speaker at technology conferences. In his spare time you can most likely find him on one of the salsa dancefloors of the world!

Abstract

On 18 November 2025, a permissions change in one internal database took down Cloudflare's Workers KV, Access, Turnstile, and Dashboard for over three hours, and a large slice of the web that sits behind Cloudflare went dark with it. Monitoring caught the spike in seconds. Finding the lead still took hours, because the real chain ran from that permissions change, through a feature file that quietly doubled in size, into the Bot Management engine, and out to every product downstream. That chain wasn't in the telemetry. It was in the topology. Most London platform teams know the smaller version of this every week: ten teams ship into one shared platform inside the same hour, the board lights up, and not one dashboard tells you which deploy lit it. The dependency graph is the data source that does. We'll walk the Cloudflare outage with a reconstructed topology in hand, next to the official postmortem. That means building the graph from what you already run, AWS, Kubernetes, Terraform, GitHub, and your monitoring, then versioning it through time, so "what changed between 14:23 and 14:47?" becomes a query instead of a guess. Once it's there, an AI agent can walk a graph it didn't invent, getting from alert to the breaking commit in seconds with the affected customers already attached. In one customer's production graph, that walk took about thirty seconds, against the twenty minutes to two hours a human familiar with the code needs to do it by hand.

Bio

Louis Fradin is a Developer Advocate & Backend Engineer at Anyshift, where he's helping build the AI context layer for production systems, giving teams the infrastructure graph they need so AI agents can actually understand what's running in prod. His path to SRE started deep in the stack: four years writing Linux drivers and managing HPC infrastructure for the French Ministry of Armed Forces, followed by three and a half years at Ubisoft building and operating Kubernetes clusters at scale for game servers with Go, Temporal, Talos, and OpenTelemetry. He speaks regularly about this work: Context Engineering for DevOps: How Agents Learn Your Infrastructure at DevOpsCon Amsterdam 2026, 25,000 servers at Ubisoft (with Vincent Behar) at TalosCon 2024 in London, Talos chez Ubisoft pour nos serveurs de jeux at Cloud Native Paris 2024, and Continuous Container Checkup at CloudNord 2023. For this talk in particular: I've spent the last year building and operating Annie, Anyshift's AI-SRE agent, and the temporal graph engine that feeds it. The Cloudflare case study is one I've walked through with our own customers on their own graphs, including the MotherDuck team, whose Head of Infra, Hilary Brennan-Marquez, put the speed-up this way: "a process that would have taken a human familiar with the code 20 minutes to 2 hours, Annie did in 30 seconds.

Abstract

Kubernetes has officially evolved from an application platform into the core operating system for enterprise infrastructure. With the rise of VMs on Kubernetes through projects like KubeVirt, SREs are now managing massive legacy virtual machines right alongside ephemeral containerised microservices and critical DevOps control planes like GitHub and Jira. When your entire engineering velocity depends on this unified stack, treating data protection as an afterthought or a manual checkbox is an operational hazard. This session outlines a modern strategy for data resilience, demonstrating how to treat your backup infrastructure entirely as code. We will walk through how to integrate immutable data protection policies directly into your declarative GitOps pipelines (ArgoCD). You will learn how to handle the massive data footprints of KubeVirt VMs, safeguard your vital DevOps tooling registries, and automate multi-cluster recovery workflows so that when disaster strikes, your state restores just as quickly as your code.

Bio

Michael Cade is a seasoned technology executive with over 20 years of experience in the data protection space. He is currently serving as the Global Field CTO for a leading cloud-native and open-source data protection company. Exploring the world of Cloud-Native and OSS software and raising awareness that there is still very much a requirement to protect data and workloads within these platforms. Prior to this role, Michael held various technical roles within Veeam joining in 2013 as a Systems Engineer and then moving to the product strategy team to focus on the wider Veeam strategy of creating content and sharing within the community. A big focus for Michael has been based on 3 things, Raising Awareness, Driving Adoption and showing the Art of the possible.

Abstract

The SDLC has compressed. Waterfall gave us months between intention and production, Agile gave us weeks. AI-assisted development gives us the cloudburst: intention becomes shipped code almost immediately. Most organisations are absorbing this flood with drainage systems designed for drizzle. This talk introduces a practical framework for AI adoption built on one governing rule: no compression capability unlocks until its matching reservoir exists. Compression (spec-driven development, agentic loops, AI-generated code at volume) creates velocity. Reservoirs (review architecture, policy-as-code, feature flags, chaos engineering) create the capacity to absorb it safely. Teams that build compression without reservoirs don't move faster; they flood. Drawing on experience shipping regulated digital identity products at Yoti, including production AI evaluation systems combining deterministic checks, LLM-as-judge, and human-in-the-loop validation, I'll close with the emerging accountability gap when AI agents act on behalf of humans, and why SREs will feel it first.

Bio

Alttaf is Director of Engineering & AI Innovation at Yoti, where he has spent nearly a decade shipping regulated digital identity products including eSignatures, verifiable credentials, and verified calls. He holds a patent in the eSignature space and has built production AI evaluation systems for regulatory monitoring. His work focuses on how engineering teams adopt AI safely: compressing delivery without outrunning their capacity to absorb it.

Abstract

This talk will focus on the latest developments affecting the SRE space, including the proliferation of AI tools and how both Platform Engineering and Observability influence SRE. There will be information from the latest events, enterprises and the SRE Report 2026. Key takeaways are, learning about the latest developments influencing the SRE space, the technologies in demand and upcoming that help SRE practitioners and strategic tips on how to navigate the high seas of supporting business services by crafting an SRE Operating Model.

Bio

Richard Simon is a CTO and a Cloud Native, Open Source and GenAI advocate with 36 years of IT industry experience, focusing on Cloud Computing, in the last 15 years. He is a highly structured, pragmatic and opinionated IT leader who's accumulated a vast set of varied experiences, working at prestigious IT vendors, Cloud Consultancies and Global Systems Integrators, such as IBM, Lenovo, SUSE, Mirantis, Heptio, World Wide Technology (WWT), Contino and T-Systems. Richard is the host of the YouTube channel, Cloud Therapist, and enjoys sharing his insights and challenging his audience with thought-provoking content, as well as during his public speaking engagements through conference keynotes and as host to various events.

Abstract

How Alibaba Cloud applies AI to real SRE workloads at cloud scale. How AI helps engineers move faster from alert to understanding: triaging noisy signals, correlating telemetry, summarising incident context, and suggesting likely causes. How AI can support root-cause analysis, capacity forecasting, incident copilots, postmortem learning, and guarded remediation without removing human accountability. How to build the foundation for AI-enabled SRE: observability pipelines, runbook knowledge bases, incident history, tool permissions, audit trails, confidence scoring, and rollback paths

Bio

Yitian Xu is Head Solution Architect for the UK&I, Nordics, and Baltic team at Alibaba Cloud, where he shapes the cloud, AI, and enterprise architecture strategies behind some of the region's most ambitious digital transformations — including key accounts such as BP, Unilever, Lego, and IKEA. With over 15 years across AI, IT, Cloud Computing, Enterprise Architecture, ITIL, Big Data, and DevOps, he sits at the intersection of large-scale infrastructure and the next wave of intelligent systems. He's here to swap ideas with builders pushing the frontier of multi-agent, real-world AI.

Abstract

I kept asking this question on our EKS clusters. kubectl top gives percentages, not dollars. AWS Cost Explorer sees instances, not pods. Every tool that connects the two wanted me to deploy Helm charts, agents, and dashboards. I just wanted a number. So I built Burn, an open-source CLI that reads your kubeconfig, fetches real-time pricing from AWS and Azure APIs, and gives you per-namespace cost breakdown in 30 seconds. No agent, no dashboard, no cluster changes. Running it on production, I found 33% idle capacity ($117/month on a 5-node cluster), a pod requesting 500m CPU but using 0.12m, and debug pods nobody remembered deploying. I deleted the waste that same day. In this talk I'll cover: - How Burn splits node cost into CPU and RAM using ratio-based pricing - Why P95 metrics matter more than averages for rightsizing - How we detect Ingress-based load balancers that other tools miss - Honest trade-offs of an agentless approach vs full platforms like Kubecost Attendees will leave knowing how to identify idle resources, understand Kubernetes cost allocation math, and evaluate the right level of cost tooling for their clusters.

Bio

Cloud/DevOps Engineer who enjoys solving complex problems and building efficient, scalable systems. My expertise includes Kubernetes, AWS, and CI/CD pipelines, with a strong focus on automating processes to simplify developers' workflows. Additionally, I maintain a supercomputer (HPC) at NEU IBM Center, ensuring it runs at peak performance for high-demand tasks. Creator of Burn, an open-source Kubernetes FinOps CLI.

Abstract

Everyone is talking about AI agents, but most organisations are still working out how to create real business value from them and, more importantly, how to operate them safely and reliably at scale. Over the past 18 months, I've worked with some of Microsoft's largest enterprise customers across engineering, construction, energy and professional services as they move from AI experimentation into production. The conversations are no longer about chatbots and copilots—they're increasingly about autonomous workflows, governed AI agents, and systems that interact with critical business processes. In this session, I'll share what customers are actually building today, where they're seeing value, and the operational challenges they're encountering along the way. We'll explore why reliability, observability, governance and operational readiness are becoming as important as the AI models themselves. Topics covered include: The shift from copilots to agentic AI Real customer use cases and lessons learned Common governance and operational challenges What SRE and platform teams need to prepare for How organisations can scale AI safely without introducing new operational risk This isn't a session about AI hype—it's a practical look at what is happening on the ground today and what it means for the next generation of enterprise workloads.

Bio

Seema Sharma is a Senior Specialist at Microsoft focused on Azure, Data and AI transformation across Microsoft's largest enterprise customers in the UK. She works with organisations to modernise data platforms, adopt Microsoft Fabric, implement AI solutions, and establish the governance and operating models required to run these technologies at scale. Her recent work has centred on helping enterprises move from AI pilots to production-ready agentic AI solutions that deliver measurable business outcomes.

Abstract

We know how to build an agent platform. The build side is staffed, the patterns are published, and the demos work. What almost nobody has solved is the run: who owns the agent on a Tuesday when its behaviour drifts, the git diff is empty, and the platform dashboard is still green? Deployed to production is not the same as approved to operate live. Between the two sit four acceptance tracks — security, data, compliance, operations — each needing a named owner, and decision latency has quietly replaced engineering as the bottleneck. This talk is an enabling session, not a gap lament. Solving for a single business unit is fine for the POC phase; scale needs an operational platform with a whole-organisation SLA — agent registry, gateway, identity, named Day-2 ownership, and a lifecycle plan for multi-agent systems that keep evolving after go-live (models improve, harnesses get tweaked, tools and data get bolted on).

Bio

Jaroslav Pantsjoha is a Google Developer Expert and Technical Director at a global consultancy, presently leading agentic AI solution delivery as Principal Solution Architect for enterprise clients in EMEA. An avid technologist with 20+ years taking infrastructure, platforms, and now AI agents from demo to production.

Abstract

Automation promises speed but speed without judgment is how a routine storage migration corrupts a live production database in minutes. While reducing EBS storage across a 142-server fleet targeting roughly 8TB of savings worth an estimated $8,000–$15,000 one database server had quietly fallen outside the disaster recovery checklist. When the automation ran, that single blind spot let a live rsync corrupt roughly 700GB of production data. This talk pulls back the curtain on the reliability engineering behind the recovery: reading metrics and logs in real time to confirm system health, why keeping the old volume until verification turned near-disaster into a clean rollback, and why the real failure wasn’t the script it was a gap in DR coverage nobody caught until production found it. You’ll leave with a principle every SRE should tattoo on their runbooks: disaster recovery has to cover every system by default, not just the ones you remembered because automation is only ever as safe as the recovery plan standing behind it.

Bio

Gabriel Okiri is a Senior Platform, DevOps & Cloud Security Engineer with 5+ years of experience spanning financial services, logistics, and AI. My work centers on infrastructure automation, cloud reliability, and recovering production systems when things go sideways because they always eventually do. I have also built AI agents and MCP (Model Context Protocol) integrations that help DevOps teams offload repetitive troubleshooting so they can focus on the judgment calls that actually matter.

Abstract

SRE has given us powerful ways to design systems that survive failure: observability, incident response, error budgets, graceful degradation and postmortems. But what about the humans responsible for keeping those systems running? Engineers can spend hours investigating a failing service while overlooking their own warning signals: reduced concentration, decision fatigue, escalating frustration or the inability to switch off after an incident. By the time the human system fails, the consequences can affect reliability, collaboration and decision-making too. This talk explores what happens when we apply reliability thinking to ourselves. Drawing on experience in software engineering, psychology and IT wellbeing, I introduce a practical Human SRE model: Observe, Respond, Recover and Learn. We'll explore how to recognise cognitive overload earlier, build recovery into the operating model, and use blameless postmortem thinking to examine human factors without turning them into individual blame. I'll also explore where AI can provide a cognitive support layer for incident preparation, reflection and decision-making. The goal isn't to make engineers tolerate more load. It's to design healthier operating conditions for the humans who keep our systems reliable.

Bio

Gena Frangina is the Founder and Host of the IT Stress Relief Podcast, exploring the intersection of technology, human performance and sustainable ways of working. With a background in software engineering, an MSc in Business Psychology and professional training in clinical hypnotherapy, she brings a cross-disciplinary perspective to stress, cognitive load and resilience in technology environments. Her work applies systems thinking to the human side of technology, exploring how professionals and organisations can create healthier and more sustainable ways of operating.

Abstract

Organisations deploying multi-account AWS environments struggle with slow, error-prone manual infrastructure deployments that take 45-90 minutes per cycle. This talk shows how to extend AWS Landing Zone Accelerator with automated CI/CD pipelines supporting both Terraform and CloudFormation, delivering rapid cross-account deployments with built-in security scanning, validation, and governance controls. You'll learn how a hub-and-spoke architecture centralises pipeline operations, implements least-privilege IAM, and maintains compliance through automated tooling. Based on a published AWS Developer Blog post and open-source solution deployed in UK Public Sector environments.

Bio

Anwaar Hussain is a Cloud Infrastructure Architect at AWS, where he specialises in architecting, automating, and developing cloud solutions with a strong focus on security compliance. He is a member of the AWS Golden Jacket UK Community and author of posts on the AWS Developer Blog, Dev Community, and Medium. He is passionate about helping organisations build secure, scalable infrastructure on AWS and contributes to open-source solutions on AWS Samples. Outside of work, Anwaar enjoys travelling, cooking, spending quality time with family, and learning new languages.

Abstract

AI systems are moving from demos into production, but operating them reliably requires more than the traditional CPU, memory, uptime, and error-rate signals. As LLMs, RAG systems, vector databases, AI agents, and GPU-backed inference services become part of real customer-facing environments, SRE teams are being asked to support a new class of workload. These systems do not always fail like traditional web applications. A request can return HTTP 200 while the user still experiences slow token generation, poor retrieval quality, broken tool calls, model rate limits, fallback failures, or unexpected cost spikes. This talk looks at AI workloads through an SRE lens. Instead of focusing on whether AI will replace SREs, it explores the opposite problem: what happens when AI itself becomes the production system that SRE teams must keep reliable? We will walk through the reliability path of a modern AI application, including the API layer, prompt gateway, model server, vector database, external model provider, GPU infrastructure, queues, fallbacks, and agent/tool-call workflows. From there, we will map traditional SRE concepts such as SLIs, SLOs, observability, capacity planning, progressive rollout, rollback, and incident response to AI-specific production risks. The session will cover practical AI workload indicators such as Time to First Token, end-to-end inference latency, tokens per second, queue depth, GPU saturation, retrieval success, fallback success, rate-limit errors, and cost per request. We will also discuss why CPU and memory alone are often not enough for scaling LLM inference, and why AI systems need visibility across infrastructure health, model-serving behaviour, retrieval quality, dependency health, and user experience. Attendees will leave with a simple production readiness framework for AI workloads: define the user journey, choose meaningful SLIs, set realistic SLOs, observe the full inference path, plan for capacity and cost, design safe fallbacks, prepare rollback strategies, and treat AI systems as critical production services rather than impressive demos.

Bio

Trupti is a Senior Site Reliability Engineer and Co-Founder of OpsCortex Labs, focused on Kubernetes, SRE, observability, cloud infrastructure, and AIOps. She creates practical engineering content that simplifies production reliability concepts using diagrams, real-world examples, and incident-focused explanations. Her work has reached 100K+ engineers across the DevOps and SRE community, with 1M+ impressions on technical content. She holds certifications including CKA, AWS, Linux, and HashiCorp, and is passionate about helping engineers understand complex production systems with clarity, strong fundamentals, and practical automation.

Abstract

Every SRE knows the feeling. An alert goes off, Slack lights up, dashboards are everywhere, and suddenly everyone is trying to answer the same questions: What changed? Is this related? What’s the safest thing to do next? We’ve become very good at detecting incidents, but responding to them is still surprisingly manual. We jump between logs, deployments, monitoring tools, CI/CD pipelines, cloud consoles, and runbooks, piecing the story together under pressure. In this talk, I’ll share why I believe the next step for incident response isn’t another dashboard or another AI chatbot—it’s systems that can investigate alongside engineers, connect the dots across different sources, explain what they found, and recommend (or even safely carry out) the right action under clear operational guardrails. We’ll look at what this future could look like, the engineering challenges behind it, and why trust, governance, and human oversight are just as important as intelligence if we’re going to automate production operations responsibly.

Bio

Paschal Ifediora is an engineer and founder of Scrubbe, an autonomous incident response platform for engineering teams. He’s passionate about distributed systems, reliability engineering, AI agents, and building systems that help engineers spend less time firefighting and more time building. His work focuses on making incident response faster, safer, and easier through intelligent automation.

Abstract

Agents now generate large, correct-enough infrastructure changes faster than any human can review them and terraform validate/plan only prove intent, not that a change is actually safe against the real world. This talk shows how we can build new harnesses that allow us to exponentially increase the velocity with which we ship code by using the paradigm of the software factory using a real-cloud digital twin that deploys every change (plan → apply → destroy against real cloud providers) before it ever reaches a production repo.

Bio

Ehsan Ashouri is a platform and infrastructure engineer at the UK Ministry of Justice.

Abstract

From Reactive to Intelligent: Site Reliability Engineering is Moving Towards Autonomy. As organisations deploy increasingly complex and distributed architectures, traditional methods of incident response based on reactive practices fall behind. The need arises for smarter automation which would allow for quicker detection of anomalies, faster context analysis, and timely interventions within defined constraints. In this sense, agentic SRE is emerging as a discipline in which AI agents learn to analyze telemetry data, establish relationships among various system components, and help SREs with triage, diagnostics, and troubleshooting while respecting certain boundaries. In this talk, I will discuss the principles of building a robust and dependable platform for agentic SRE, the key components that constitute the architecture, and the observability layer which provides foundational data. I will also examine the necessary safeguards required to ensure autonomy while at the same time keeping operations secure and auditable. Finally, through examples of practical use cases, such as alert triaging, incident copiloting, root cause analysis, and bounded automated remediation, we will explore how AI can be integrated into operations to improve reliability.

Bio

I help platform engineers and DevOps teams understand and adopt cloud-native infrastructure through talks, demos, and community building. I organise the CNCF Gandhinagar, HashiCorp UG Gandhinagar, and Google Cloud Gandhinagar communities, and I have spoken at 40+ events, including KubeCon India, PlatformCon, HashiTalks, OpenSearchCon Korea, FOSS ASIA, and LinuxFest Northwest. My core focus areas are Kubernetes, GitOps, IaC with Terraform/Vault, AI-powered DevOps tooling, and observability pipelines. I've mentored 15+ hackathons and open-source programmes. Certifications: GCP DevOps Professional, Azure AZ-104, FinOps Certified Engineer, Terraform Associate, Vault Associate.

Abstract

For three years I was the only platform engineer at a security startup, and honestly, I didn't really know what "SRE" meant while I was doing it. There was no team, no on-call rota, nobody else to page , i.e. just me and whatever was broken. This talk is about the decisions I made with no one around to tell me if I was making the right call, and what that does to you. Over time, moments like that turned into habits I didn't plan to build , such as tearing down environments every night, making every pipeline safe to reverse, treating every deploy like it might be the one nobody else can fix. I don't think this is how anyone should learn reliability. But it's how I learned it, and I think there's something useful in seeing what those instincts look like when nobody hands them to you.

Bio

Saurabh is a Senior DevOps/Platform Engineer has worked across with multiple teams including small startups to giant banking firms. At heart, he's a learner and tinkerer who loves DevOps and building things, has a soft spot for startups, and is also a teacher.

Abstract

Before your first container hits production on Kubernetes you install a distro, a CNI, an ingress, cert-manager, Prometheus, Grafana, Loki, ArgoCD and Harbor. None of them is your app. 57% of Kubernetes users run more than 11 separate components, the control plane is a pet that needs restoring from backup at 3am, and "why can't A talk to B?" sends you through DNS, Endpoints, kube-proxy, iptables, CNI logs and NetworkPolicy.

I have been evaluating, running and being paged for Kubernetes since 1.0 in late 2015, and I kept notes. This year I stopped complaining and built the thing: Reliaburger, a container orchestrator in Rust that ships scheduling, gossip membership, Raft, eBPF service discovery, ingress, mTLS, an image registry, metrics, logs, GitOps and chaos testing in one binary, driven by TOML. An app with health checks, ingress and autoscaling is twelve lines. There is no overlay network, no CNI and no kube-proxy: a name, a virtual IP and about 390 lines of C in the kernel.

The talk covers the four scars and what Reliaburger does about each: apps instead of pods, a control plane that is elected rather than installed, SWIM gossip on one slide, and one connection from web to redis step by step. It also covers what it cost to build with Claude and Codex (200 hours and about a thousand pounds for roughly 220k lines of Rust), the rules that kept the models from doubling the codebase every week, the design we got wrong, and what Reliaburger deliberately does not do. Then a live demo, if the demo gods allow.

It is 0.1.0, free and Apache 2.0. Download it, break it, tell me.

Bio

Miko Pawlikowski has been running platform teams on top of Kubernetes since version 1.0, and has the scars to show for it. He is the author of Chaos Engineering (Manning) and of Reliaburger, a batteries-included container orchestrator written in Rust, along with the book Building Reliaburger that documents how every subsystem was designed and built.

Abstract

AI agents generate thousands of investigations in parallel, changing the economics of storage and compute. Discover why security and observability platforms are evolving toward open storage, elastic compute, and decoupled architectures.

Bio

Peter Marshall is an award-winning speaker, technology leader, and community builder with 25 years' experience in data architecture and digital transformation. As Director of Developer Relations at Imply, he leads programs that grow and engage global communities through education, support, and events. With experience across startups, enterprises, and the public sector, Peter brings technical expertise and strategic vision to help organizations leverage real-time data and observability technologies. He holds a BA in Theology and Computer Studies from the University of Birmingham

Abstract

Self-service infrastructure makes it easy for humans and agents to provision databases without involving SRE or platform teams. But when things go sideways, those same teams can find themselves on-call for infrastructure they didn’t provision. This session explores how teams can maintain resilience as database estates grow, and how an agentic operations layer can reduce the operational burden by continuously observing infrastructure, and recommending remediations and optimizations across the estate. It looks at how teams can move from scaling incident response with database count to policy-based operations, where agents and humans work together, with humans governing what actually executes.

Bio

Rob Reid is a Technical Evangelist at Cockroach Labs and an experienced software engineer with a background spanning backend, frontend, messaging, and distributed systems. He is the author of Practical CockroachDB and has also written books on CockroachDB and multi-region application architecture.

Abstract

Every talk about AI agents begins after the interesting part. The agent is already connected: it holds a credential, it can reach the database, and the discussion starts from what it does next. But that secret is the least examined object in modern infrastructure, and in almost every production system running today, it does not belong to the agent at all. It is borrowed. A service account, a pipeline credential, a static string inherited from whatever the application authenticates as. The access decision is not about the agent. It is about a costume the agent is wearing.And this is not only a production concern: the credentials that matter most often sit in build pipelines and evaluation environments, where they are least governed and most useful to an attacker.

This talk argues that every agent must be a first-class principal: an identity in its own right, cryptographically attested, scoped to a single task, and expiring on its own. We will look at how long-lived credentials actually fail in practice: stolen, half-rotated, or simply never revoked, and why each of those is an identity failure rather than a model failure. We will examine the assumptions underpinning conventional identity systems that autonomous actors break, why machine identities now outnumber human ones by two orders of magnitude while every governance control we have built still protects the human, and what any credible identity model must answer: who is this actor and for how long, what may it do and can that authority only narrow, and what did it do and on whose behalf. Prompts are not permission boundaries. Identity is.

Bio

Olu is a Solutions Architect at Teleport, where he helps organisations solve one of the hardest problems in security: giving every human, machine, workload and AI agent a cryptographically secured identity. He works with engineering and security teams to design and operationalise that shift end to end at scale, drawing on a background in building secure, resilient cloud-native and hybrid infrastructure. Olu has delivered webinars and spoken at conferences, sharing his perspective on securing identity and access across increasingly complex and heterogeneous environments.