{"free":true,"mirrors_paid_route":"GET /api/dossier?slug=ai-agents-tech-stack-deep-dive-2026-09","price_if_paid_usd":"0.25","slug":"ai-agents-tech-stack-deep-dive-2026-09","title":"AI Agents — Technology Stack Deep-Dive","skill":"deep-dive-dossier","format":"markdown","excerpt":"# AI Agents — Technology Stack Deep-Dive\n\n**Research date:** September 2026  \n**Addendum to:** [AI Agents — Deep-Dive Dossier](./ai-agents-deep-dive-2026-09.md)  \n**Audience:** Advisory clients evaluating build vs buy and stack choices  \n**Scope:** Public documentation and attributable vendor claims only; extends (does not contradict) the parent dossier’s Technology stack section\n\n---\n\n## Scope & how to read this\n\nThis addendum unpacks the parent dossier’s **Technology stack** section into decision-grade detail: what to optimize for in models, which runtime/SDK to pick, when to buy a platform control plane vs build on open frameworks, how MCP/A2A change integration risk, and what “good” looks like for memory, evals, and AgentOps.\n\n**How to use it**\n\n| If you need… | Jump to… |\n| --- | --- |\n| Model selection criteria (not brand) | Models for agentic workloads |\n| SDK / runtime comparison","truncated":true,"headings":["Scope & how to read this","Models for agentic workloads (what actually matters beyond brand; long-horizon, tool use, cost, caching)","Agent runtimes & SDKs (OpenAI Agents SDK, Anthropic Claude Agent SDK, Google ADK, Microsoft Agent Framework — capabilities, when each fits)","Low-code / platform control planes (Copilot Studio, Agentforce, Vertex/Agentspace, Bedrock Agents — SoR fit)","Open orchestration frameworks (LangGraph, CrewAI, LlamaIndex, Pydantic AI, Mastra, etc. — production readiness notes)","Tooling protocols: MCP in depth (architecture, auth/enterprise concerns, Shadow MCP, AAIF; practical adoption checklist)","Agent-to-agent: A2A (role vs MCP; maturity)","Memory & context systems (short/long-term patterns; failure modes)","Eval, tracing & AgentOps (what “good” looks like in production)","Reference architectures (2–3 diagrams in mermaid or clear ASCII: single-agent + tools; multi-agent; platform-native)","Build vs buy decision matrix","Stack recommendations by client situation (3–4 archetypes)","Sources"],"full_bytes":33519,"note":"Free inspectable preview — full markdown is behind the paid fetch route.","buy_url":null,"parked":true,"status":"PARKED","hero_paid":"GET /api/enrich","note_buy":"PARKED — not for sale. Use GET /api/enrich ($0.02) or GET /api/tip ($0.01)."}