CTO PulseAI

Highlights From Software Architecture Superstream: Software Architecture and the Age of Agentic AI

O’Reilly Media
O’Reilly Media•May 4, 2026

Why It Matters

Agentic AI can reshape development pipelines, but only with robust architectural controls will enterprises reap productivity gains without compromising quality or governance.

Key Takeaways

  • •Agentic AI demand surges, outpacing generative hype momentum.
  • •Code reviews risk quality when AI handles massive commits.
  • •Data contracts, traceability, and governance are core architectural pillars.
  • •Semantic layers provide context to prevent AI hallucinations in queries.
  • •Context engineering expands prompt design into multi‑modal, event‑driven workflows.

Summary

The Superstream session examined how software architecture must evolve for the age of agentic AI. Speakers highlighted the rapid rise in interest for autonomous AI agents, noting that while generative models have matured, the ability of agents to act independently is now the primary focus for developers and enterprises. Key points included the danger of relying on AI for code reviews of oversized commits, the current capability gap in building reliable micro‑services, and the necessity of data contracts, traceability, and governance as foundational architectural attributes. The discussion also introduced semantic layers as a solution to provide the contextual knowledge agents lack, and described context engineering as an expansion of prompt engineering into multi‑modal, event‑driven pipelines. Illustrative examples featured a developer’s claim that AI‑driven code reviews are faster—only because commits become too large to review manually—and a scenario where an agent is asked for Q3 revenue without fiscal calendar context, leading to hallucination. The panel also described agents interacting with different LLMs for text, voice, or image tasks, coordinated through event‑driven communication. The implications are clear: organizations must adopt rigorous data contracts, enforce traceability, and build semantic layers to harness agentic AI safely. Without these architectural safeguards, AI‑augmented workflows risk errors, compliance breaches, and reduced software quality, ultimately affecting business outcomes.

Original Description

The evolution of agentic AI for software architecture is moving faster than most teams can track, and the people best equipped to cut through the noise are the architects and engineers who've spent careers designing systems. In this highlight reel from O'Reilly's Software Architecture Superstream, Neal Ford, Sam Newman, Pramod Sadalage, Mary Grygleski, and Cornelia Davis tackle the hard questions practitioners are facing right now.
Sam Newman challenges the claim that AI is speeding up code reviews, arguing those reviews aren't better, just faster, because AI-generated commits are so massive that no one is actually reading them. Pramod Sadalage gets concrete about what happens when an agent lacks organizational context (it hallucinates your Q3 revenue figures because it has no idea your fiscal year starts in February, not January), then makes the case for semantic layers as the fix. Mary Grygleski maps out multi-agent architectures where specialized LLMs handle text, voice, and images, all stitched together through event-driven design. If you're designing systems where agentic AI plays a role, this is a conversation you don't want to miss. Cornelia Davis rounds things out by reframing the conversation entirely: Context engineering isn't just a fancier name for prompt engineering. It's the broader discipline of crafting and maintaining the rich context that makes agentic systems actually trustworthy.
Follow O'Reilly on:

Comments

Want to join the conversation?

Loading comments...