Sources
All accessed 2026-08-19. [vendor] marks vendor-published claims about the publisher's own market.
Wang, Zhu, Feng, Lu and Jia, agentic AI and human-in-the-loop interventions at Alibaba customer service (arXiv 2605.14830; experiment Aug 2024, paper May 2026): the randomised field experiment anchoring supervision load, escalation mix, and the specialisation question. Bainbridge, Ironies of Automation (Automatica, 1983). Crandall and Cummings, performance metrics for supervisory control of multiple robots (HRI 2007): the fan-out measurements. Cummings and Mitchell, predicting controller capacity (IEEE SMC-A, 2008): the wait-time overestimate. Whetten, Goodrich and Guo (IEEE SMC 2010): deliberate slack beats maximum fan-out. Endsley and Kiris (1995), Endsley and Kaber (1999), Endsley, Ironies of artificial intelligence (Ergonomics, 2023). Parasuraman and Riley (1997); Parasuraman and Manzey (2010, a narrative review, not a meta-analysis); Lyell and Coiera (JAMIA, 2017). Bucinca, Malaya and Gajos, cognitive forcing functions (CSCW 2021). Vasconcelos et al., explanations can reduce overreliance (CSCW 2023). Bansal et al. (CHI 2021). Becker, Rush, Barnes and Rein, METR randomised trial (arXiv 2507.09089, Jul 2025). Chen et al., comparing human oversight strategies for computer-use agents (arXiv 2604.04918, Apr 2026). Green, flaws of policies requiring human oversight (Computer Law and Security Review 45, 2022). Sterz et al. (FAccT 2024). Elish, moral crumple zones (2019). McCarley et al., deconstructing the vigilance decrement (Attention, Perception and Psychophysics, 2026; preregistered, N=366). Humlum and Vestergaard (RFBerlin DP 078/26, Mar 2026). Brynjolfsson, Chandar and Chen, Canaries in the Coal Mine (Aug 2026 version). Hosseini and Lichtinger (Oct 2025). Audoly, Guerin and Topa, NY Fed (May 2026). MIT CISR AI maturity work (2025). Wharton and GBK (Oct 2025). Pew Research Center (Oct 2024). Markoff, Skilled Work Without the Worker (New York Times, Aug 2012) for the metaphor correction. EU AI Act Articles 14 and 26(7). EEMUA 191 third edition and 10 CFR 50.54(m) for consequence-weighted supervision floors. Microsoft Entra ID Governance documentation and Anthropic autonomy measurement and containment posts (2026) [vendor]. Workforce law and consultation evidence: EU AI Act Articles 4, 14, 25(1)(c) and 26(7); AI Omnibus Regulation (EU) 2026/1744 (in force 27 Jul 2026) deferring Annex III high-risk to 2 Dec 2027; European Commission second-stage social partner consultation on the Quality Jobs Act, SWD(2026) 239 final (20 Jul 2026), for the adoption, representation, information and contestability statistics and the named legal gaps; OECD employer survey on algorithmic management (2025); ILO Working Paper 144, global case studies of social dialogue on AI (Jul 2025); European Parliament resolution 2025/2080(INL) on algorithmic management (adopted 17 Dec 2025); Platform Work Directive (EU) 2024/2831, Articles 7 to 13, transposition due 2 Dec 2026; Betriebsverfassungsgesetz sections 80(3), 87(1) No. 6, 90(1) No. 3, 95(2a) and 111 as amended in 2021; ArbG Hamburg 24 BVGa 1/24 (16 Jan 2024); BAG 1 ABR 20/21 (8 Mar 2022); CJEU C-634/21 SCHUFA (7 Dec 2023); German Datenschutzkonferenz guidance on AI and data protection (6 May 2024); Hans-Boeckler-Stiftung Mitbestimmungspraxis Nr. 59 (Feb 2024) for the named company agreements; Eurobarometer 554 and the European Working Conditions Survey 2024. Do not use, verified wrong or untraceable: the "128 robots, 9 workers" figure; "entry-level hiring down 80%"; any specific agent-to-operator ratio; "agentic AI skills postings jumped 986%". Re-verify by Q1 2027: the labour-market contest; whether any credible supervision ratio is published; EU workforce-consultation enforcement practice.
Source: research/R13-operating-model/sources.md in the evidence repository behind this site.