Resources — high-trust sources for Domain Agent Design

Ground every non-obvious claim in these; cite inline (<sup> → #sources). Populated before teaching. Prefer primary sources (vendor engineering, papers) over secondary commentary. This track is design discipline, so several sources are internal Engineering Vault tracks the learner has already built — cross-link, don’t re-teach.

Primary specs / docs

• Anthropic — “Building effective agents” (anthropic.com/research/building-effective-agents). Why it matters: the spine for when to build an agent at all. Workflow vs agent, “start with the simplest thing,” agents trade latency/cost for autonomy. Every domain blueprint starts here.
• Anthropic — “Writing effective tools for AI agents” (anthropic.com/engineering/writing-tools-for-agents). Why it matters: tool surface design — how many tools, naming, error shapes. Directly feeds the “tool surface” and “responsibility boundary” parts of each blueprint.
• Anthropic — “Effective context engineering for AI agents” (anthropic.com/engineering/effective-context-engineering-for-ai-agents). Why it matters: the context layer each domain agent depends on; cross-links to ../context-engineering.
• Anthropic — Claude tool-use / Claude documentation (docs.anthropic.com). Why it matters: the concrete tool-definition + tool-result loop that every domain agent runs on.
• OWASP — Top 10 for LLM Applications (owasp.org). Why it matters: blast radius and approval-gate design. Excessive agency, prompt injection, and insecure tool use are the failure modes a domain agent’s boundary must contain.

Papers & deeper reading

• Yao et al., 2022 — “ReAct: Synergizing Reasoning and Acting in Language Models” (arXiv:2210.03629). Why it matters: the reason-then-act loop underneath every domain agent; grounding actions in observations is what makes a domain agent more than a chatbot.
• Engineering Vault — ../ai-agents (internal track). Why it matters: prerequisite. LLM, tool, agent, agent loop, runtime, memory, MCP. Lesson 0004 (0004-agent-systems-and-production.html) introduces business agent archetypes; this track turns that into a repeatable design discipline.
• Engineering Vault — ../context-engineering (internal track). Why it matters: prerequisite. How each agent’s context is selected/retrieved/assembled — the layer this track designs around, not in.
• Build-vs-buy / SaaS framing (general engineering literature; e.g. Martin Fowler on capability vs commodity, martinfowler.com). Why it matters: most domain-agent decisions are buy-the-platform, build-the-domain-glue. The ROI and build-vs-buy parts of each blueprint lean on this.

Notes on trust

• Vendor blogs describe their product’s current strategy; treat product specifics (model names, limits, exact tool conventions) as version-dependent and verify against current docs before teaching as fact.
• ROI and cost numbers are illustrative. Teach the shape of the trade-off (cost per action vs human cost, deflection rate, blast radius) and hedge concrete figures as “at time of writing.”
• The two internal tracks are the source of truth for shared terms — reuse their definitions verbatim so vocabulary never drifts across tracks.