AI Agents in Depth boundary

Type: agentic-system-analyses/types/agentic-system-boundary.md

Boundary and evidence

The target is the book's shipped companion agent workflows and operating guidance, not an asserted single production runtime. The explicitly defined class agentic workflow and teaching suite means a collection of separately launched model-dependent experiments, with shared packaging and explanatory material. The intended use is to distinguish those implemented mechanisms from the book's recommendations and reproduction plans. It is in scope because shipped drivers issue model calls through agent implementations; the manuscript alone is not being classified as an executing agent. SRC-1: docs/en/README.md, pyproject.toml, chapter1/context/main.py, chapter1/context/agent.py.

Agent = LLM + Context + Tools — This book builds on this core formula across 10 chapters, taking AI Agents from principles to engineering practice. The full text, illustrations, and 93 accompanying experiments are all open source. You are welcome to run the experiments yourself. --- docs/en/README.md

The project count is the source's declaration, not an independently verified count of working runtimes. The root package describes shared experiment plumbing, rather than supplying evidence of a unified scheduler or shared state loop. SRC-1: pyproject.toml.

description = "Shared packaging and plumbing for the ai-agent-book companion experiments" --- pyproject.toml

Initial entry-point tracing followed the README's chapter1/context/main.py example to its ContextAwareAgent import, CLI modes, ablation-suite construction and the implementation's model request. This establishes a shipped model-dependent path, not successful execution or task correctness. SRC-1: docs/en/README.md, chapter1/context/main.py, chapter1/context/agent.py.

Call the model with tools

response = self.client.chat.completions.create(**create_kwargs) --- chapter1/context/agent.py

A second trace inspected the user-memory driver's construction of a conversational agent and background processor, then followed the consumer's disk reload and prompt construction and the processor's history reload and analysis-agent handoff. These shipped callers belong to the target, not to an excluded host. This preliminary trace establishes that the suite includes state maintenance and later consumption, not merely a memory API. It does not establish complete coverage of its memory modes or evaluators. SRC-1: chapter3/user-memory/main.py, chapter3/user-memory/conversational_agent.py, chapter3/user-memory/background_memory_processor.py.

self.memory_manager.load_memory()

Add memory summary

memory_str = self.memory_manager.get_context_string() --- chapter3/user-memory/conversational_agent.py

Analyze conversation - this now directly updates memories via agent tools

The agent will process the conversation and use its tools to update memories

_ = self.analyze_conversation(conversation_context) --- chapter3/user-memory/background_memory_processor.py

Functional scope and inspection limits

All tracked companion implementations, their shipped drivers, prompts, maintenance procedures, evaluators, fixtures and retained experiment artifacts at SRC-1's commit are eligible evidence. The following inventory separates initial inspection from material still requiring inspection. It is not a whole-system coverage claim or a path allowlist.

The boundary kind is complete artifact, partial loop, not whole-system. The Git artifact is frozen in full, but several reproduction paths depend on other repositories and external environments not frozen here. The root documentation explicitly distinguishes standalone projects, reproductions and designs and requests separately obtained external repositories. Its appendix also retains historical chapter-path labels; it is not evidence that those components are present or executable at the current tree paths. SRC-1: docs/en/README.md, pyproject.toml.

The 23 external repos for benchmarks, training frameworks, and robot platforms in Chapters 6, 7, 9, 10 are not bundled (due to size and licensing) and must be cloned into the corresponding directories. --- docs/en/README.md

No external repository, linked paper, live website, provider response, model weight set, benchmark download or separately installed host is registered. Tracked vendored content, if present, remains eligible as part of this Git artifact, not as a claim about a separately acquired upstream revision. Missing external components prevent end-to-end conclusions about reproduction and training loops. External model services and host environments prevent conclusions about service behavior, deployment and host instruction loading. Available but unread files are inspection limits, not access gaps. SRC-1: docs/en/README.md, pyproject.toml, skills/README.md.

The frozen revision is dbc046eb896ac4e39aa19c7774c8bf49583b89a6; checkout HEAD was verified against acquisition, with no working-tree changes reported. The applicability cutoff is 2026-10-12, the supplied run date. It is not a promise about later main-branch or release contents. Overall evidence is code-grounded, limited to source inspection. No target code, test, example or model call was executed. Retained traces may support later observed-run findings only after their contents and conditions are inspected; they do not automatically raise this boundary's evidence layer.

Source register