Sources
All accessed 2026-08-19. [vendor] marks vendor-published claims about the publisher's own market.
Context engineering genealogy: Lutke (Jun 19 2025), Karpathy (Jun 25 2025), Willison (Jun 27 2025), Anthropic (Sep 29 2025) [vendor], Gartner Innovation Insight (2025); OHR-Bench (arXiv 2412.02592, ICCV 2025); Docling (arXiv 2408.09869; LF AI & Data); Anthropic contextual retrieval (Sep 2024) [vendor]; late chunking (arXiv 2409.04701); semantic-chunking cost question (arXiv 2410.13070, NAACL Findings 2025); NVIDIA chunking benchmark (Jun 2025) [vendor]; MMTEB (arXiv 2502.13595); Embedding-Converter (ACL 2025); S3 Vectors GA (Dec 2025) [vendor]; CoALA (arXiv 2309.02427); MemGPT/Letta; Zep (arXiv 2501.13956) and Mem0 (arXiv 2504.19413) [both vendor-authored, disputed]; LangMem (Feb 2025) [vendor]; sleep-time compute (Apr 2025) [vendor]; AgentCore Memory GA (Oct 13 2025) [vendor]; AgentPoison (NeurIPS 2024); MINJA (2025); Microsoft agentic failure taxonomy (Apr 2025, updated Jun 4 2026); vec2text (EMNLP 2023; reproduction arXiv 2507.07700); OWASP LLM08 (2025); Azure document-level ACLs (2025) [vendor]; Restricted SharePoint Search (Apr 2024); Concentric oversharing figures (secondary, moderate confidence); "When More Documents Hurt RAG" (arXiv 2606.11350); coverage-trust trade-off (arXiv 2607.05217); ICLR 2025 passage-count degradation; Moffatt v. Air Canada (BC CRT, Feb 14 2024); LinkedIn ticket KG (arXiv 2404.17723); KCS v6 roles; O'Reilly year-of-LLMs canon (2024); LLM-judge curation (arXiv 2601.17717; audit reduction arXiv 2607.22766). Excluded as unverifiable: circulating citation-accuracy statistics. Re-verify by Q1 2027: memory-service dispute state; many-to-one ACL patterns; context-engineer role consolidation.
Source: research/R14-agent-data-engineering/sources.md in the evidence repository behind this site.