Sources
All accessed 2026-08-19. [vendor] marks vendor-published claims about the publisher's own market.
Datadog State of AI Engineering 2026 (Jul 2026, Mar 2026 telemetry, 1,000+ organisations, dependency-inspected rather than self-reported); Sierra Ventures CXO survey (Jul 2026); Menlo Ventures State of Generative AI in the Enterprise (Dec 2025); Anthropic building effective agents (Dec 2024), multi-agent research system (Jun 2025), context engineering (Sep 2025), effective harnesses for long-running agents (Nov 2025) and harness design (Mar 2026) [vendor]; Cognition, don't build multi-agents (Jun 2025); Tran and Kiela, budget-controlled single-versus-multi comparison (arXiv 2604.02460, Apr 2026); Cemri et al., MAST failure taxonomy (arXiv 2503.13657, NeurIPS 2025); Kulkarni and Kulkarni, financial document orchestration patterns (arXiv 2603.22651, Mar 2026); PerspectiveGap orchestration prompting benchmark (arXiv 2606.08878, Jun 2026); AssetOpsBench retrospective (arXiv 2605.08518, May 2026); Lu et al., The Routing Plateau (arXiv 2606.07587, May 2026); Kang and Wang, registry-governed agent lifecycle (arXiv 2607.00345, Jul 2026); Yi and Song, harness-induced belief divergence (arXiv 2607.04528, Jul 2026); Sen et al., agent harnesses and agentic search (arXiv 2605.15184, May 2026); Cao et al., skill marketplace analysis (arXiv 2607.09065, Jul 2026); IMDA Singapore Model AI Governance Framework for Agentic AI (Jan 2026) with its named enterprise case studies; Meta engineering, ranking engineer agent (Mar 2026); Microsoft Agent Framework harness and declarative workflows posts (Jul 2026) [vendor]; Temporal durable-execution guidance (Jul 2026) [vendor]; Gartner agentic project cancellation forecast (Jun 2025). Corrections made during research, do not repeat: a widely circulated single-versus-multi-agent token comparison attributed to a benchmark paper is a misreading of that paper's domain fingerprint and sample size; and the claim that the leading anti-multi-agent essay preceded the leading pro-multi-agent post by three months is contradicted by the publication dates, which are one day apart. Unverified, do not cite: a specific compaction token-reduction percentage; a router cost-reduction headline; a skills benchmark pass-rate improvement; framework maintenance-mode support dates. Re-verify by Q4 2026: framework consolidation state; whether the multi-agent research figures are updated; lifecycle patterns moving from proof of concept to observed operation.
Source: research/R07-agent-platform/sources.md in the evidence repository behind this site.