RAG-over-vector broke at scale on relational reasoning, temporal coherence, and state-machine reasoning. The hybrid three-layer architecture is now the production default at JPMorgan, Pfizer, Adobe, and SAP.
Eight frontier models on five capability axes, three operational axes, and two regulatory axes. The procurement-grade decision is no longer a ranking; it is a matrix, and cost-per-useful-output is the decisive metric.
2.1 million accelerators on one Iowa campus, the Gemini 3 training runs that landed there, and the JAX-versus-PyTorch ecosystem split.
Microsoft, Oracle, Meta, and Crusoe at 191,000 units committed. ROCm 7, the TSMC commitment, and the inference-versus-training split.
Munich Re publishes MR-SAUC 1.0 — the first cross-insurer standard for AI-augmented underwriting — with Swiss Re, Allianz, AXA, and Zurich as co-signatories.
BCG rebuilt its proposal pipeline on Claude and a 30-year archive. The win rate is up 18 points. The Vienna conference disclosed the operating data.
Pictet, UBS, JPM, Morgan Stanley, Goldman. Five portfolios decomposed, sleeve by sleeve, fee by fee.
RAG-over-vector broke at scale on relational reasoning, temporal coherence, and state-machine reasoning. The hybrid three-layer architecture is now the production default at JPMorgan, Pfizer, Adobe, and SAP.
Eight frontier models on five capability axes, three operational axes, and two regulatory axes. The procurement-grade decision is no longer a ranking; it is a matrix, and cost-per-useful-output is the decisive metric.
Morgan Stanley Wealth Management, AIA Group, and the Generalitat de Catalunya signed three Cohere contracts worth between $148M and $190M committed. The graded-jurisdiction positioning is winning.
Optum, ING, McKinsey, and FedEx ran four parallel agent-stack consolidations. The pattern is not vendor convergence; it is category-redistribution along the dominant-constraint dimension.
2.1 million accelerators on one Iowa campus, the Gemini 3 training runs that landed there, and the JAX-versus-PyTorch ecosystem split.
Microsoft, Oracle, Meta, and Crusoe at 191,000 units committed. ROCm 7, the TSMC commitment, and the inference-versus-training split.
256GB unified memory, 76 TOPS NPU, and the Fortune 500 pilots that mark the first workstation-class on-device LLM inference at scale.
A field study of 612,000 Blackwell units shipped in Q1 2026 — hyperscalers, neoclouds, enterprise direct, and the secondary-market response.
Munich Re publishes MR-SAUC 1.0 — the first cross-insurer standard for AI-augmented underwriting — with Swiss Re, Allianz, AXA, and Zurich as co-signatories.
BCG rebuilt its proposal pipeline on Claude and a 30-year archive. The win rate is up 18 points. The Vienna conference disclosed the operating data.
A field study of Fortune 500 AI procurement spend in Q1 2026. Vendor consolidation, multi-vendor strategy, and the shift from pilot to production budgets.
Twelve months after creating the Chief Intelligence Officer role, Deloitte's review documents the org chart, the budget authority, and the four named programs.
Pictet, UBS, JPM, Morgan Stanley, Goldman. Five portfolios decomposed, sleeve by sleeve, fee by fee.
Hebbia, Glean for Finance, AlphaSense-Tegus, plus the editorial-as-research layer. The new UHNW research stack.
$42B across eight sovereign funds. The picks-and-shovels positioning, decomposed.
Vontobel, EFG, and Julius Bär scale tokenised private credit through SDX. The DLT Act becomes operational.
Four hundred twelve contested procurement processes. Granola 58, Fireflies 26, Otter 16. The buyer-side rubric the category trade press keeps inverting.
Eighty-four operators surveyed across six months. The four-part format the cohort converged on, the 0.7-point decision-quality differential, the failure modes the practice will need to survive.
Forty-seven migrations tracked. Seven named. The typed-schema advantage agent stacks read for context, and Atlassian's narrowing strategic response.
A Series B software company, a 300-person consultancy, a 1,000-person media operation. Sixty-eight Agents in production, three failure modes the buyers actually log.
Forty Fortune 500 organisations now run a discrete agent-platform team. JPMorgan, Walmart, Bank of America, Workday, Salesforce. The reporting line and budget authority shape the function's posture.
Hatchpay's 80-engineer organisation moved from 2.8 calendar days median bug-fix to under 0.7. The agent alone produced 1.5×. The Linear integration and hooks-based policy enforcement produced the rest.
LangGraph and CrewAI stayed. AutoGen forked into AG2. OpenInterpreter and Plandex folded. The dependency graph tells the truer story than the GitHub stars.
Vercel v0 holds 64 per cent of Next.js scaffold deploys. The deploy funnel, the AI SDK distribution, and the framework's evolution toward v0's patterns explain the concentration.
Twelve months after the Vaud filing, the function is no longer experimental. CHF 18M operational envelope. Three programs live. 23 per cent cross-network transaction lift. The role is now structural, and the sector is watching.
Twenty maisons surveyed. Three group doctrines, four independent postures, one deep secret. The parent dynamic now predicts deployment architecture more strongly than any technical or commercial variable.
Axel Dumas does not deploy AI in the work of the artisans. Hermès deploys AI everywhere else. The internal taxonomy of operational versus creative AI organises every investment decision the Maison has made.
Claude on the CRM. Cohere on the archive. ElevenLabs on VIP voice. Fifteen flagships from Place Vendôme to Ginza. Eleven per cent lift in UHNW average transaction value. Nineteen per cent more bespoke commissions through Métiers d'Art.
BfArM, AIFA, HAS, AEMPS — four competent authorities, four postures. The EU AI Act enforcement phase begins 2 August. The transatlantic gap with FDA has widened, and it is structural.
AMA's CPT Editorial Panel finalises a Category III code for diagnostic-agent augmented clinical reasoning. CMS proposes $4.21 per encounter — cents on the dollar against the E/M code, but it adds up.
First cross-border export of Cleveland's ACIS framework into Europe. Three institutions across Germany, Sweden, Austria. The framework held; the local modifications illuminate what depends on what.
NEJM AI publishes the field's first prospectively-designed multispecialty diagnostic-agent outcome study. 47,318 encounters, four specialties, four-arm comparison. The effect sizes are modest, consistent, and large enough to matter.
Multi-source investigations. Sourced, scored, audited.
Same-week notes. Declarative. Operator-grade.
Inside the room. Named operators on background.
Every byline names a model. Every editor is disclosed. Coverage is governed by a Swiss-law charter that names what we will and will not do.
