Software Architecture Superstream: Building Architectures for Agents (and Humans)

Building architectures for AI requires software architects to structure information so that AI models deliver consistent, high-quality results. You want to build a reliable bridge between human goals and machine execution, ensuring that the AI functions as a predictable part of your overall system.

Neal Ford et al. explore the new toolkit software architects need as AI becomes a fixture in many systems. They also discuss the techniques you can use to communicate intent to AI, including architecture as code, and look at the trade-offs you must consider. Whether you’re curious about how to manage complex AI workflows or just want to make your tools more dependable, this event provides a practical roadmap for building smart, stable, and scalable AI systems.

What you’ll learn and how you can apply it

• Spot the failure modes that only show up in production, including evaluation debt, silent semantic failures, and agents that die when infrastructure resets
• Get architectural patterns for governing nondeterministic systems, such as deterministic boundaries, runtime intervention, and fitness functions that don’t rely on the model behaving itself
• Walk away with frameworks you can use: a maturity model for evals, a containment model for agent authority, and five questions for deciding what agents may propose versus what needs independent approval

Recommended follow-up

• Read Architecture as Code (early release book)
• Read AI Agents: The Definitive Guide (book)