Architecting Production-Ready Gen AI and Agentic AI Systems
A Comprehensive Practitioner Playbook
A practitioner playbook for the architecture decisions behind production GenAI and agentic systems, from decision contracts to durable execution and governance.
- Author
- Srinivas Bommena
- Edition
- Version 3
- Published
- April 2026
- Formats
- Kindle
- Length
- 536 pages
About this book
Version 3, August 2026.
An AI system becomes an architecture problem the moment its output can change a workflow, invoke a tool, influence a consequential decision, consume governed data or keep acting after the original request has ended.
At that point, model quality is only one part of the design. The architecture must also decide what the system may know, which sources it may trust, what authority an agent inherits, where deterministic validation sits, how long-running work survives failure, what happens when a tool call partially succeeds, and what evidence will let the organisation reconstruct the decision months later.
The book begins before model selection, distinguishing AI-added, AI-first, generative, agentic and hybrid systems and connecting them to organisational maturity and the level of autonomy the enterprise can actually operate. It then develops a canonical production architecture covering decision and data contracts, retrieval and knowledge systems, prompt and context engineering, evaluation, agent runtimes, multi-agent orchestration, deployment, resilience, security, FinOps, observability and operational governance.
What’s new in Version 3
- Agentic AI as a first-class design concern: single-agent and multi-agent architectures, delegated identity, bounded authority, durable execution, state, retries, approval boundaries and trajectory-level evidence.
- A stronger production runtime: synchronous, streaming, queued and durable execution; backpressure and event-driven decoupling; and the Propose → Validate → Commit pattern that separates probabilistic reasoning from deterministic business execution.
- Agent-aware evaluation, security, observability and governance, with the NovaCred case connecting architecture decision records, tool manifests, release gates, cost baselines, runbooks and approval trails into one evidence model.
Who it’s for
Solution architects, AI engineers, platform teams, engineering leaders, product owners, risk professionals and senior practitioners designing enterprise GenAI and agentic AI systems.
