AI agents can generate code, but generating code is only a small part of working safely with real software systems. To be useful in production, agents need architectural context, reliable documentation, observability, clear boundaries, and access to the right domain knowledge.
This talk explores how engineering teams can prepare existing systems for effective collaboration with AI agents without sacrificing reliability, security, maintainability, or human judgment. We will look at practical architectural patterns for making systems easier to understand and navigate, including explicit module boundaries, deterministic workflows, structured documentation, meaningful observability, and guardrails around automated changes.
The session will also discuss what AI agents should and should not be trusted to do, how to expose system knowledge safely, and how to evaluate AI-assisted changes through testing, review, and operational feedback.
Attendees will leave with a practical framework for assessing whether their systems are ready for AI-assisted engineering, along with concrete steps to improve architecture, documentation, and developer workflows before introducing more automation.