As AI coding assistants and autonomous agents become standard across the software development lifecycle, engineering organizations face a critical question: how do we meaningfully measure developer productivity? Traditional vanity metrics like lines of code, pull request counts, or the percentage of AI-generated code often create perverse incentives without reflecting actual business value or software health.

This hands-on, interactive workshop equips engineering leaders, platform engineers, and practitioners with a structured method to define metrics that actually matter. Participants will deconstruct high-level goals into clear concepts, input drivers, and measurable proxies. We will explore how to balance delivery throughput, system stability, cognitive load, and user-centric outcomes in an AI-assisted engineering environment.

The Kubernetes audit log is one of the most underutilized data sources in platform engineering. Traditionally focused on security compliance, it holds a second largely untapped layer of signal, one that captures every behavioral pattern that determines whether your Internal Developer Platform is succeeding or silently failing.

Every time a devops engineer bypasses the standard pipeline, executing directly into a pod, forwarding a local port, or applying hotfixes outside the GitOps flow, the audit log records it. These events are the early warning signs of platform friction and tool abandonment. Almost nobody is listening to them.

This session shows how to turn those raw audit log entries into actionable platform intelligence. We walk through the exact architecture required to capture behavioral patterns and translate them into a unified data model. We break down five critical bypass patterns like kubectl-exec, port-forward, pipeline-skip, manual-merge, and config-drift, and demonstrate how to compute a decay-weighted Friction Score per service, surfaced directly inside Backstage, the open-source framework powering developer portals at over 1,000 organizations worldwide. This score provides a continuous, quantitative answer to the question every engineering leader asks: "Is our platform actually working, or are teams fighting it?".

Audience Takeaways:

• Capture only the audit log events that signal a platform bypass with no extra cost and no alert fatigue.

• The 5 Bypass rules: Field-level JSON mappings for kubectl-exec, port-forward, pipeline-skip, manual-merge, and config-drift. All are ready to implement in any audit pipeline.

• A decay-weighted model that scores platform health continuously, self-correcting as team behavior improves.

• Convert raw bypass metrics into executive dashboards that replace gut feeling with engineering-grade evidence.

The activities of software development have barely changed. Requirements, design, build, test, operate: the same five have appeared in every model since the 1950s, from Waterfall through Agile to DevOps. What changed each time was not the work but who does it, in what order, and alongside whom.

AI agents are the next shift of that kind. They now write code, generate tests, produce pipelines, and propose fixes. But which of these activities should they lead, and which still belong to us? How do we decide, and on what basis? Most teams answer case by case, and the inconsistency shows up as agents drifting without direction, outputs nobody fully trusts, and changes nobody can explain after they ship.

This workshop introduces the Human–AI Lead Map, a taxonomy that places every activity in the delivery lifecycle according to who leads it. It covers software development practices and extends across DevOps and DevSecOps: pipelines, infrastructure, security gates, observability, and incident response. Through engaging exercises, attendees will place real activities on the map themselves and leave with a clear view of where that line sits in their own organization.