Longer-form work a board paper, procurement file or literature review can cite. Every piece states its sources in full, and every framework carries a DOI.
Six versioned frameworks and one reference architecture. Each has a diagram, a suggested citation and a PDF brief deposited on Zenodo with its own DOI.
01Five-Dimension AI Readiness AssessmentFramework and PDF brief →
02TRACE FrameworkFramework and PDF brief →
03Four-Question AI Governance BaselineFramework and PDF brief →
04Return on Employee (RoE) FrameworkFramework and PDF brief →
05Measurement-Before-Prediction ArchitectureFramework and PDF brief →
06Governed Cognitive Layer (Layer 8)Framework and PDF brief →
ArchitectureGoverned Agentic RAGRead the architecture →The Readiness Assessment, TRACE, the Governance Baseline and the architecture's control set are also published as YAML, templates and a JSON Schema in the open ai-governance-toolkit repository under CC BY 4.0.
Field notes record failures observed in the AI systems that run this site, each with its root cause, the control added and the ISO/IEC 42001 clause or Annex A control it maps to.
Each case study follows one organisation or policy over years, built from annual reports, regulator records and named press sources, with every claim numbered to its source. Some are public; client briefs are password protected.
Tom Blomfield’s May 2026 YC talk argues companies should run as AI loops that fix themselves overnight, with no middle managers and every conversation recorded. Each of its ten main claims set against published research, agent incidents and the law, with a map of which loops can run unattended and the controls a board needs for the rest.
Singapore produces about 790,000 tonnes of food waste a year and recycles 18% of it. The evidence for AI in predicting and preventing waste, improving collection and recovery, and building measurable value; who is doing what across government, industry and the community; where Singapore universities are active and where their work is thin; and the routes for engaging them.
A neutral case study of the most highly rated AI programme in banking. What JPMorgan did well: data and cloud foundations first, one governed generative AI platform for more than 200,000 staff, senior owners and named operating results. What it has not shown: a published method for its “about $2 billion of benefit”. Its statements on jobs, the data-control penalties of 2022 and 2024, how DBS measures AI value by comparison, and lessons for boards.
Five families sued Character.AI over harm to their children. What the Florida court decided at the pleading stage and what it did not, the undisclosed settlements, how the company changed its product, the first companion chatbot laws in New York and California, and what boards deploying conversational AI should build in from the start.
How Klarna’s claim that its AI assistant did the work of 700 agents was read as 700 jobs lost, changed across its SEC filings, and drew a question from the regulator after the CEO spoke of lower quality: the headcount and audited cost record, and what boards should ask before publishing AI productivity claims.
How Commonwealth Bank of Australia declared 45 call-centre roles redundant after a voice bot went live in 2025, then said 23 days later they were not: the union dispute, the apology, the cuts that followed in 2026, and what boards should ask before acting on an AI system's effect.
How a fraud crackdown, a self-learning risk model that used nationality and an all-or-nothing recovery rule harmed tens of thousands of Dutch families, 2013 to 2026: the regulator's findings and fines, the cabinet resignation, a EUR 11.8 billion recovery, and the lessons for AI.
Australia's automated welfare debt program, 2015 to 2020: how income averaging created about 794,000 debts without a legal basis, the warnings that went unheeded, A$1.751 billion written off, the accountability that followed, and the lessons for automated and AI decisions.
What stops an AI agent from moving money it shouldn't: the scam and deepfake evidence, the MAS, IMDA and CSA expectations, ISO/IEC 42001, and four questions answered by seven controls that produce evidence a board and an auditor can inspect.
Fifteen years in sequence: insourcing and fixing the basics, the 2014 digital vision, the 2018 move to 33 two-in-a-box platforms, the data and AI platforms, the 2021 to 2023 outages and MAS actions, and what the S$1 billion of economic value does and does not measure.
Where Kedai Mesra and Café Mesra lose sales and margin between the pump and the shelf, ten agentic AI opportunities, and a pilot plan. Case study and lecture deck.
Thirteen concepts, each with a working proof of concept you can use: plain requests in, real actions out, with consent, policy and an audit trail around every step. Password protected. Ask for the password to try them.
AI broker listing and proposal portalOpen the project →
Bank of the FutureOpen the project →
CTCB AI: a voice assistant for security officersOpen the project →
AI cargo planning for transport aircraftOpen the project →
Agentic AI internal audit systemOpen the project →
Agentic gold trading deskOpen the project →
Conversational sales desk for WhatsApp, Telegram and emailOpen the project →
Agentic recruitment operations with every hiring decision kept humanOpen the project →
Agentic talent sourcing for a recruitment agency and its bank clientOpen the project →
Agentic tour operations for a destination management companyOpen the project →
Purchase-with-purchase offers at the fuel pumpOpen the project →
Agentic partner desk for industrial connectivityOpen the project →
Carpark access and reporting for a car showroomOpen the project →One study, a doctoral thesis and two books, with the numbers and chapters as they come in. Password protected because the work is preliminary. Ask for the password to follow it.
Your sample is smaller than you thinkOpen the study →
The Governed EnterpriseOpen the book page →
Sustained AI adoption in Singapore SMEsOpen the thesis page →
NexusMind: agentic AI on premisesOpen the brief →
Sovereign AI data centre, MalaysiaOpen the brief →
Project Permittiv: radar that screens treesOpen the brief →
Sustainability data and building intelligenceOpen the master plan →
AI at Scale: From Pilot to ProductionOpen the book page →The longer articles, rebuilt as technical reports with an abstract and a numbered reference list, and deposited on Zenodo with a DOI. Eight are co-written with domain leaders in finance, cybersecurity and fuel retail. Practitioner reports, not peer reviewed.
With Dr. Christina Chua
With Chris der Kinderen
I built this page for readers who need more than a blog post. A board paper, a procurement file or a literature review needs something it can cite and check, so the frameworks carry DOIs and every case study numbers its sources. The case studies follow real organisations over years, from their own filings and their regulators' records, because the order in which an organisation made its decisions teaches more than any announcement. I will keep adding deeper research here. If one of these helps you win an argument you should have won anyway, that is what it is for.
