Weekly to fortnightly
Briefings are sharper, single-question pieces between 1,200 and 2,500 words. Published weekly to fortnightly.
The most consequential layer of the AI buildout is not the foundation models themselves, but what sits between them and the organisations that deploy them: architecture, integration, evaluation, and governance. This is the applied layer — and the public record of the past year has clarified its shape rather than settled it.
Central proposition. By 2026 the frontier models have converged on capability for the median enterprise workload, and the evidence reviewed here indicates that production AI quality is shaped more by architectural composition than by model selection. The same model in a well-architected system and in a naive pipeline produces materially different outcomes.
Central proposition. The evidence reviewed here converges on the operating model — rather than technology choice, including the now-live choice between Western and Far East frontier models — as the stronger predictor of enterprise AI outcomes. This reading is consistent with two decades of IT-economics research establishing that technology investment produces returns only alongside complementary organisational investment (Brynjolfsson & Hitt, 2000; Tabrizi et al., 2019).
The headline cost of model inference is a small and shrinking fraction of what enterprises actually spend to run generative AI in production. The production evidence reviewed here — modelled from public pricing and vendor disclosures rather than audited enterprise ledgers — indicates inference accounting for 20–40% of run-rate cost for mature deployments. Retrieval, evaluation, observability, governance, and human review consume the remainder — and are largely invisible at planning time.
Enterprise AI systems fail in production in characteristic ways. Across the cases reviewed here — Air Canada, Avianca, iTutorGroup, State Farm, DPD, Chevrolet of Watsonville, all of which entered the public record through litigation or press — the recurring commonality is the absence of two operational disciplines: evaluation (the ongoing measurement of whether systems perform as intended) and governance (the delivery of policy as code, controls, and accountable workflows).
The five-pillar research programme, the stakeholders each pillar serves, and the objective IDs that drive the evidence base.
How the aims, objectives, and questionnaire translate into evidence, findings, decisions, and a forward research agenda.