Method. 5 search angles → 24 sources fetched → 110 claims extracted → top 25 verified via 3-vote adversarial checking (≥2/3 refutes to kill) → 19 confirmed, 6 refuted → synthesized. 106 sub-agents, ~3.2M tokens.
Time-sensitivity is high. Figures pinned to late-2025 / early-to-mid 2026 (MCP ~97M downloads ~Mar 2026; LangChain survey Nov 18–Dec 2 2025; Gartner 40%-by-2026 projection).
Sources. Primary: arXiv 2505.03864 · OpenTelemetry blog · LangChain State of Agent Engineering + LangGraph 1.0 docs · Gartner press release · O'Reilly "AI Agents Stack 2026." Secondary/vendor (directional): MLflow · Kellton · Redis · Vectorize · Langfuse · Galileo · Inngest · framework-comparison blogs. Full URL list preserved alongside the source report.
Confidence tags reflect verification status at research time, not eternal truth. Verify vendor/blog-sourced claims before acting on them.