AI Agent Infrastructure
An Independent Engineer's Guide to Building the Platform Behind Production Agents: Runtimes, Model Gateways, Memory, Sandboxes, Observability and Scale
-
- $0.99
-
- $0.99
Publisher Description
A practical, independent guide to the infrastructure that production AI agents run on, written for platform engineers, SREs, cloud architects and AI engineers. Building one agent is a project; running dozens of agents for many teams is a platform problem. Following the agent platform team of a fictional global e-commerce retailer, the book designs a complete reference architecture and explains each layer with worked numbers: compute and runtimes from serverless to Kubernetes, durable execution engines for long-running agents, queues and events, model gateways with routing, fallbacks, rate limits and tenant quotas, self-hosted inference with vLLM-class servers and GPU sizing, session and memory stores on Postgres, Redis and vector databases, MCP server hosting and tool registries, sandboxed code execution with containers, gVisor-class kernels and microVMs, workload identity, delegated authorization, secrets and egress control. It then covers observability with OpenTelemetry GenAI conventions, evaluation pipelines, CI/CD and infrastructure as code for agents, FinOps for tokens and GPUs, reliability and capacity planning, multi-region deployment, data residency under several regions' laws, and build-versus-buy decisions. Includes checklists, a lab, study questions and a glossary.
This book is an independent publication and is not authorized, sponsored or endorsed by any company whose products it mentions. All trademarks belong to their respective owners and are used only for identification. This book was drafted with generative AI tools and checked with automated tools. Tools, features and prices change often; confirm them in current documentation.