What Volume I is
Volume I (May 2026 – May 2027) is a literature and landscape synthesis. The five flagship reports are built from public primary sources: vendor technical reports and arXiv preprints, regulatory primary text (EU AI Act, Korea AI Framework Act, China’s CAC Interim Measures, India’s MeitY guidelines and DPDPA, Japan’s METI/MIC AI Guidelines and AI Promotion Act, NIST AI RMF and Generative AI Profile, ISO/IEC 42001:2023, FDA AI/ML guidance), named-enterprise case studies (DBS, ING, MUFG, Sanofi, Bayer, JPMorgan, Klarna, Stripe, Rakuten, Wiz, Mayo Clinic, Walmart, Tesco, Bloomberg, Schneider Electric, H&M, Hitachi, Bharti Airtel, Reliance Jio, Sarvam, Naver, Sakana, and others), official benchmark submissions (LMArena, SWE-bench Verified, BEIR, MTEB, MIRACL, OpenCompass, C-Eval, GAIA, OSWorld), and methodologically-transparent industry surveys (McKinsey State of AI 2025, MIT NANDA GenAI Divide, Stanford AI Index 2025, BCG Where’s the Value in AI?, BoE-FCA AI survey).
What Volume II adds
From Q3 2026 the interview programme activates: structured practitioner questionnaires, Patron-only roundtables, and longitudinal cohort tracking of named programmes. Field notes from inside real deployments enter the mix as relationships allow. The research base shifts from synthesis-of-public-evidence to a hybrid that includes first-hand primary signal. Volume II in May 2027 delivers the publication’s first annual synthesis, drawing on both bodies of evidence.
Source-tier rubric
| Tier | What counts |
|---|---|
| Tier A – primary | Vendor technical reports, arXiv preprints and peer-reviewed papers, regulatory primary text, official benchmark submissions, court decisions, SEC filings, model cards. |
| Tier B – direct or near-direct observation | Enterprise engineering blogs, on-the-record practitioner publications and conference talks, investor and shareholder materials, named-individual technical writing. |
| Tier C – reputable secondary press | Reuters, Bloomberg, FT, Nikkei, Caixin, The Information, BBC, Guardian, Wired. Used for context only, never load-bearing. |
| Tier D – methodologically-transparent surveys and analyst work | Flagged inline; survey n, geography, and field date disclosed. |
Western press characterisations of non-Western model labs and enterprises are explicitly Tier C and not load-bearing – direct lab technical reports and engineering disclosures are preferred.
Evidence-pairing rules
Every load-bearing claim cites at least one Tier A or Tier B source. Vendor self-benchmarks are flagged inline (for example, Voyage Rerank-2’s 93-dataset internal study). Independent reproductions are privileged where they exist (for example, GraphRAG-Bench against the original GraphRAG). Contested regulatory interpretation is flagged rather than resolved.
Geographic and editorial balance
Across the five flagship reports, approximately 38–40 percent of cited sources are headquartered outside the United States, with substantial Asia-Pacific (China, Korea, Japan, Singapore, India), European, UK, and Middle Eastern representation. The publication does not describe non-Western enterprise programmes in deficit terms relative to Western templates, and refuses any single regulatory regime as the global default.
Editorial-original frameworks
Several frameworks in the flagship reports are inductively derived editorial syntheses, not validated taxonomies. They include Pillar 1’s five-component decomposition of the applied layer (architecture and retrieval, orchestration and agentic patterns, operating models and program design, economics and platform choice, evaluation and governance), Pillar 1’s five-level whole-program maturity ladder (Exploratory, First Production, Multi-System Discipline, Production-Grade Program, Cross-Domain Compounding), Pillar 1’s seven cross-cutting principles, and Pillar 1’s six-ecosystem global view (Western US/UK/EU, Chinese, Indian, Korean, Japanese, and Middle Eastern, with European model-lab activity beyond the US-aligned stack treated as a distinct strand); the operating-model 2x2, the Six Conditions of Success, and the Operating-Model × Success-Condition Interaction Matrix (Pillar 3); the eight-layer cost stack and five workload archetypes (Pillar 4); the three governance archetypes and the five-level operational maturity framework (Pillar 5); and the five-tier agentic maturity taxonomy and six anti-patterns (Pillar 2). These frameworks are inductively derived from named-enterprise evidence, original to this publication, and not yet validated longitudinally. The interview programme that activates from late 2026 is what will test them.
Pillar 1’s maturity ladder describes whole-program AI deployment maturity (Exploratory through Cross-Domain Compounding); Pillar 5’s five-level operational maturity framework describes governance and evaluation discipline specifically (Ad-hoc through Adaptive). They measure different things and should not be conflated.
Banned vocabulary
Responsible AI, trustworthy AI, ethical AI, and AI values do not appear as load-bearing terms in this publication. They name destinations, not roads. The publication describes governance through the operational practices that translate principle into shipped control: model registry, evaluation gates, human-in-the-loop checkpoints, observability, incident response, data and identity controls, decommissioning. Hype vocabulary (revolutionary, game-changing, transformative, cutting-edge, AI-powered) is similarly avoided.
Conflicts of interest
The Applied Layer has no commercial relationship with any vendor, model lab, cloud provider, data platform, or evaluation tooling company discussed in any piece – no paid engagement, sponsorship, referral fee, equity, or revenue-share arrangement. No vendor was given pre-publication review, fact-check rights, or the opportunity to respond before publication of the flagship reports. Author relationships are disclosed in-piece. Vendor briefings, where they have informed thinking, are disclosed.
Corrections
Material errors are corrected promptly and visibly. A correction is added at the foot of the piece, with the date and a brief description of the change. Substantive editorial revisions – not typographical fixes – are surfaced in the piece’s metadata via a “Last revised” date. Pieces are not silently re-edited after publication.
Citation
Pieces published by The Applied Layer are intended to be citable. Each piece carries a stable URL and a published date. Reference Architectures additionally carry a version identifier suitable for inclusion in architecture board papers and internal standards documents. Citation in the form “The Applied Layer (YEAR), Title, URL” is welcomed.
Ongoing-research roadmap
- May 2026 – Volume I launches: five flagship reports and weekly briefings.
- Q3 2026 – Interview programme activates; structured questionnaires open.
- Q4 2026 – First Patron-only practitioner roundtable; first longitudinal cohort report.
- May 2027 – Volume II annual synthesis (the first annual synthesis): literature update plus first wave of primary-data findings; re-cut of all editorial-original frameworks against fresh evidence.