Building Production-Grade AI Agents
By Choral Software
Language: English
Reliable Runtimes, Secure Tool Use, Context Engineering, and Evaluation
Engineer AI agents that can be trusted with real work.
Building Production-Grade AI Agents is a systems engineering guide to building AI agents that can be trusted in production: reliable runtimes, secure tool use, context and memory engineering, evaluation, and operations, with a free, runnable reference implementation.
What’s inside
- Part I, Agent Systems Foundations: from LLM calls to stateful agents, autonomy boundaries, runtime anatomy, and threat and failure models.
- Part II, Context Engineering and Memory: token budgets, working, episodic, semantic and procedural memory, retrieval and reranking, compression, and permission-aware retrieval.
- Part III, Reliable Agent Runtimes: state machines, durable execution, idempotent tool calls, retries and circuit breakers, sagas, human approval, replay, tracing, and cost controls.
- Part IV, Agent Security: prompt injection, tool and memory poisoning, delegated authorization, capability-based access, secrets, sandboxing, malicious tools, and kill switches.
- Part V, Evaluation and Production Operations: trajectory evaluation, property-based testing, fault injection, red teaming, monitoring, incident response, and rollout.
- Appendices: a production-readiness checklist, a companion API reference, templates and runbooks, an index of worked examples, and case studies.
Who it is for: software engineers, platform and security teams, and technical leads who are moving AI agents from demos to systems that must be reliable, secure and observable.
Features: 44 chapters in five parts; a free, runnable companion reference implementation; checklists, templates and runbooks you can adapt to your own systems.
Availability: Coming soon.