As coding agents increasingly produce production code, the critical engineering problem shifts from writing code to supplying, controlling, and evolving context: When agents are the developers, context becomes the primary interface through which intent, constraints, and system knowledge are expressed. And luckily we have lots of context : our coding standards, the delivery requirements, feature stories but also security guidelines. With standards such as Agent.md and Agentskills.md the industry is slowly working on a more structured and interoperable way of working across all tools.

This talk presents an emerging agent-centric context SDLC, supported by a new class of Agent enablement platforms. We’ll break down concrete context primitives like specifications, documentation, rules, skills, and historical decisions and how new standards help in making them reusable, composable, and version-controlled which directly affects agent reliability, consistency, and throughput. Rather than focusing on code reviews, teams begin to engineer context inputs that drive predictable outcomes across asynchronous agent workflows : if we can improve and correct context every next code generation can be improved.

We’ll dive into how the feedback cycle works in this model. Instead of testing only code, teams introduce context evals: structured tests that verify whether agents receive the right context at the right time, and how changes in context impact correctness, rework, and cycle time. These evals form a loop that allows developer and platform teams to improve agent performance systematically. It’s time we use engineering practices to go beyond vibe coding

The talk also covers the underlying coding infrastructure and tools required to support this shift: cloud-native, event-driven, asynchronous systems that handle context distribution, execution isolation, guardrails, security boundaries, observability, and cost controls. We’ll discuss how guardrails move upstream into context definition and access, and how security concerns surface at the level of context, not just code execution. Your security and platform teams also have a lot of context to contribute!

Finally, we’ll look at coding agent enablement as a parallel to developer enablement. What practices transfer, what breaks, and where a new operating model is required? We’ll share early signals and correlations between context quality, agent efficiency, and delivery metrics. We’ll how these may start to map to familiar indicators like DORA. This talk is a practical exploration of how platform teams can enable agents and the teams of developers around delivering software more effectively.