Picking Your Winner in the Agentic AI Race

• Evaluate agents against real organizational work, not generalized benchmark standings.
• Treat identity, permissions, context, observability, evaluation and human intervention as parts of the agent platform.
• Select agents by workload instead of assuming one agent or model will win every event.
• Build portability and interoperability into the platform before agent dependencies become deeply embedded.
• Prepare platform engineering to become the discipline that connects agent capabilities to governed enterprise action.

Agentic AI is rapidly transforming DevOps—from code generation to incident response—but enterprises face a critical challenge: how to adopt these systems without introducing new risks. This session presents a practical, production-focused framework for evaluating and implementing agentic AI in enterprise DevOps environments.
Drawing from real-world architecture and release governance practices, we explore how to assess agentic systems across key dimensions such as security, reliability, cost, and operational fit. The session also introduces actionable guardrails, including role-based access control, approval workflows, auditability, and release readiness validation, ensuring that autonomous actions remain controlled and compliant.
Attendees will gain a clear approach to classify agent behaviors, define accountability metrics, and integrate AI safely into CI/CD pipelines without compromising stability or trust. The goal is to move beyond experimentation and enable confident, scalable adoption of agentic AI in enterprise DevOps.

Platform engineers are at the center of the enterprise AI shift, but only if the platform is ready for it. In this talk, Luca Galante and Kaspar von Grünberg introduce the Agentic Development Platform (ADP): a reference architecture built around three planes — tooling, path specifications, and agent infrastructure — that gives AI agents the structure they need to actually deliver value in production. Drawing on the concepts from the book Thinking in Platforms, the session shows platform teams how to evolve golden paths into agent-ready specifications and position their platforms, and their organizations, to thrive in the agentic era.

DevOps culture built on human collaboration, shared ownership, and psychological safety faces existential evolution as AI agents automate 30% of engineering workflows by 2030. Core cultural norms around team accountability, knowledge sharing, failure tolerance, and innovation velocity must adapt to autonomous systems that generate code, self-heal pipelines, and make production decisions without human intervention.

Coming soon!

AI agents are rapidly evolving from prototypes to production, and it's important to implement structured lifecycle management to ensure reliability, governance, and long-term sustainability. This session will focus on a reference architecture for designing and orchestrating GitOps-driven continuous delivery pipelines to provision, update, and govern Declarative AI Agent infrastructure end-to-end for agent configuration as code, versioning, automated validation, and safe rollback. The talk will discuss how declarative manifests, pull-based reconciliation, and environment-based promotion patterns can enable reproducible agent builds, governed updates, and auditable change history. The pipeline architecture integrates GitOps controllers such as Argo CD and AI agent frameworks such as LangGraph to standardize agent build and deployment workflows, and LangSmith for debugging and monitoring agents. Attendees will also learn about drift detection, policy guardrails, observability signals, rollback strategies, and lifecycle controls required to safely manage behavioral change, reduce deployment risk, and improve reliability, security, and governance for production-grade agent ecosystems.

As AI accelerates software delivery, it also expands risk, complexity, and attack surfaces. Traditional DevSecOps, while essential, can no longer keep pace with AI-driven systems and evolving threats.

This session introduces Intelligent Continuous Security (ICS) as the next evolution. ICS embeds security across the entire value stream, using Generative AI and Agentic AI to continuously assess risk, enforce policy, and respond in real time.

AI is both the challenge and the solution. It drives unprecedented speed in code generation and system change, while enabling autonomous security agents that monitor behavior, detect vulnerabilities, and initiate remediation at machine speed.

Attendees will learn how to move from reactive, pipeline-based security to proactive, adaptive defense systems. The session presents a practical model for implementing ICS, along with real-world scenarios that illustrate how AI augments security teams without replacing human judgment.

Key insights include how to scale security with AI, reduce emerging risks, and maintain trust through continuous validation and governance.

Security is no longer a gate.

It is an intelligent, continuous system designed to evolve as fast as the software it protects.

Release engineering is evolving beyond traditional automation to address the growing complexity, scale, and speed demanded by modern DevOps environments. As delivery pipelines span multiple tools, teams, and platforms, engineering leaders face persistent challenges around predictability, risk management, and release confidence. "Smarter, Adaptive, and Predictable Release Engineering—Reimagined with AI" explores how artificial intelligence can fundamentally transform release pipelines from rule‑based automation into intelligent, self‑learning systems.

By embedding AI across the release engineering lifecycle, teams can continuously analyze code changes, pipeline signals, test outcomes, and operational feedback to anticipate risk, optimize decisions, and improve flow. Machine learning models enable adaptive quality gates, early anomaly detection, and predictive insights that help teams identify potential failures before they reach production. AI‑driven intelligence also reduces manual triage, shortens feedback loops, and improves coordination across development, testing, and operations.

The result is a release engineering approach that is not only faster, but also more resilient and reliable. This session highlights how AI empowers DevOps teams to move from reactive releases to proactive, data‑driven delivery—unlocking greater confidence, consistency, and control in an increasingly complex software delivery landscape.

Everyone is racing to pick the winning AI agent. Almost no one is asking what the horse is running on. An agent that acts on unverified pipeline state is indistinguishable from an agent that hallucinated that state. It may win, it may bolt into the parking lot, and you will not know which until the damage is done.

This session introduces proleptic events: delivery events that carry their authored expectation forward, validated and signed before any agent is permitted to act on them. The trust model inverts. Instead of hoping the agent behaves, the substrate guarantees that everything the agent sees has been confirmed against an expectation a human authored, and that every action it takes is traceable back to that expectation.

Using Conduit, the CDF's open-source orchestration platform, we will show an agent operating inside a safe execution space, consuming only validated, signed events, producing a verifiable record of every step. The result is not a smarter agent. It is a track the race can safely be run on; the difference between autonomy you can audit and autonomy you have to apologize for.

You didn't build your AI stack. It's being built right in front of you, model by model, framework by framework, orchestration layer by orchestration layer, until the thing you now operate is one you never actually chose. And it's still forming while you're making the calls that matter: where to run it, how to govern it, what to build versus buy. Those decisions are setting defaults you'll fight to unwind later. So the question is simple: does your stack serve your goals, or have your goals started bending to fit the stack? This session shows you where lock-in hides, how observability and control get traded for speed, and how to tell a stack that compounds value from one that compounds cost and risk. You'll leave able to weigh the stack you have against the one you'd build on purpose.

When we assessed our enterprise cloud estate, the findings were sobering: a 22% Microsoft Secure Score, over 96% of workloads unprotected, privileged access management installed but never configured, and an undersized network firewall carrying the entire security burden. This talk is the honest story of turning that around across a 14-subscription, dual-region Azure platform alongside a production AWS environment.

I will walk through how we rebuilt the platform on a hub-and-spoke architecture with infrastructure as code (Bicep and Terraform), replaced credentials with managed identities and OIDC everywhere, enforced private endpoints and centralized DNS, and layered AI-assisted threat detection on top of native telemetry. I will also cover the organizational side: how we unblocked RBAC bottlenecks between deployment pipelines and security teams, and how we made secure defaults cheaper for developers than insecure shortcuts.

Attendees will leave with a practical maturity roadmap: what to fix first when your score is low, which controls deliver the most risk reduction per engineering hour, and how to measure progress in a way executives understand. This is a case study in security transformation done by a small platform team, not a big-budget program.

Most teams that add AI scanning to their pipeline quickly face more findings than they can review. That part isn’t surprising. What is surprising is what happens next: remediation gets faster and teams close findings at record speed, yet the number of unresolved critical vulnerabilities continues to climb—sometimes tripling in a year.

It’s a growing disconnect between speed and security, and most dashboards aren’t designed to reveal it.

In this webinar, I’ll unpack the industry data behind this trend, explain why faster remediation doesn’t necessarily mean lower risk, and show you how to spot the warning signs before they surface in a breach report.

Token anxiety is real because running advanced AI is core to business operations, Can your business manage the cost and supply-chain risk of depending entirely on Frontier LLM providers? Leading enterprises driving a shift to hybrid AI so they can ensure access and manage costs of AI workloads. However, the cost and complexity of running your own systems can be overwhelming.

In this session, we'll review how leading operators are delivering self-managed AI infrastructure at speed and scale using heterogenous platforms. We have practical advice on how you can succeed and what traps to avoid. We'll cover what actually changes when you move from hosted LLMs to bare metal, including the lifecycle management, patching cadence, and process controls that separate teams who deliver from teams who drown. You'll leave with a practical framework for evaluating self-hosted AI and what it takes to run it well.