What if I told you that agents can produce secure, stable software?

According to the 2025 DORA report, AI-driven development is increasing throughput, but also increasing instability. In fact, a major blocker to adoption concerns security and quality issues. The result is often "slop," increased technical debt, and out-of-control batch sizes (like single commits with 1,802 deletions) that bypass review and break builds. In the worst cases, agents have deleted or leaked codebases.

This talk demonstrates how to mend that instability without killing the speed of agentic coding. We will move beyond basic Context Engineering using a single file to applying DORA best practices and security policies before a single line of code is written. You will learn how to enforce a strict Red/Green/Refactor loop that gets coding agents to work in small, known-good increments rather than generating massive, untestable blobs. Then you'll see how to add security as early as possible to mitigate the majority of issues before a team needs to worry.

Finally, we will show "shifting left" with agents using tools like security extensions and automated checkpoints to catch issues such as injection vulnerabilities, misconfigured IaC files, and business logic issues before they even commit to your repo.

You will leave this talk able to build software with an agent that follows test-driven development, extreme programming, and proactive application security.