Our consultants are in client businesses every week. These are the arguments, patterns and practical methods that come out of that — written for people who have to make a decision on Monday, not for people collecting reading material. Fourteen articles so far, seven more already scheduled.
Agentic business, ontology, engineering, governance, learning
CIOs, CDOs, L&D leaders and architects who have to choose
Singapore and Hong Kong, written with global reference
Already written and scheduled; they publish as they clear review
If you read nothing else, read these two. They carry the argument the rest of this site rests on.
Using AI is not the same as being agentic. A definition built on three necessary conditions — shared business semantics, bounded authority, measurable output — and the six layered terms that get confused with each other every week. Includes how human and machine get governed as one class of operating asset.
Nine testable conditions a platform must meet before it is allowed near production data — each scored from evidence the vendor produces in front of you, never from a demonstration. Plus the disqualifiers that should end an evaluation before a pilot ever starts.
Our own coverage map. If a topic is in a pale column we will say so rather than pretending to expertise we do not have — which is also why this page is not a list of everything anyone has ever published about AI.
| Agentic business | Agentic operations | Ontology & data | Context engineering | Governance | Enterprise learning | |
|---|---|---|---|---|---|---|
| Definition and framing | ||||||
| Comparison and selection | ||||||
| Implementation guidance | ||||||
| Governance and evidence | ||||||
| Workforce and skills | ||||||
| Regional focus |
Filter by the area you are working in. Everything published is here — nothing is held back for the newsletter.
A six-dimension scoring method for candidate processes, the T0–T4 autonomy tier table, three scenarios to avoid, and a 90-day plan that produces a number finance will accept.
What agentic sales really means: AI agents hold defined authority over qualifying, researching, drafting, following up and reconciling, while a named human stays accountable for the pipeline. Which four stages to rebuild first, and why busier representatives is the symptom of doing it wrong.
How enterprise ontology differs from a glossary, a data model and a knowledge graph — and why the tenth use case costs a fraction of the first. With the six semantic layers and a reuse economics table.
Singapore enterprises are usually not short of pipelines — they are short of shared definitions. The four-layer estate, why semantic layers fail without named ownership, and what agentic workloads change about the economics of getting it right.
Not a battle for the standard retrieval layer — a mapping from question type to retrieval shape, applied to your own workload. Cost and latency, indexing overhead, and the hybrid setups that actually get deployed in production.
The three ways RAG fails, a ten-parameter retrieval tuning checklist with what happens when you move each one, and when GraphRAG earns its cost — and when it does not.
The most scalable evaluation method available today and the easiest to misuse, because a judge returns a confident number whether or not it is measuring anything. Rubric design, pairwise comparison, failure modes, and the calibration that keeps scores honest.
Usually framed as a choice, and it is not. An API is the contract one system calls another with; MCP is the protocol that lets an agent discover and use capabilities consistently. Which layer you expose, and the security differences that matter when you do.
ISO 42001 against ISO 27001, NIST AI RMF and the EU AI Act. Plus the governance problems traditional IT controls do not cover — autonomy, tool calls, cross-system action and unpredictable output.
Each system owns a different question. Without a single employee record underneath, they drift. Why adding an AI layer on top of three disconnected systems makes the problem worse, and how to sequence it properly.
Most enterprises hold the pieces and cannot produce the chain — and the gap shows up at audit, when there is no time to rebuild it. The six-link evidence chain, the record fields auditors ask for first, and retention periods that hold up under external review.
Most programmes deliver content and measure attendance, then discover that nothing changed. The gap is not curriculum quality — it is that the old workflow was never redesigned. Capability baselines, practice over content, cohort design and measuring behaviour.
Coming in the next batch. Proof of concept versus diagnostic: when you are buying a pilot instead of an answer · Digital employee onboarding: how agents actually join an org chart · The Human Machine Capital framework · AI governance requirements in Hong Kong SAR · RAG tuning parameters: a hands-on guide · Getting AI cost into the finance conversation · Partner enablement frameworks for external ecosystems. Subscribe to the briefing and we will send them as they publish.
Because the argument only holds if all the parts connect. An agentic business without ontology is a demo. Ontology without learning is a model nobody operates.
Operating models, work decomposition, authority allocation, digital workforce strategy.
Business semantics, semantic layers, knowledge graphs, enterprise data engineering.
Retrieval, context engineering, agent architecture, evaluation, observability.
ISO 42001, AI risk classification, autonomy tiers, audit evidence, controls.
LMS, talent management, competency frameworks, reskilling, evidence and compliance.
Instructional design, custom content, simulation-based learning, authoring at scale.
External partner enablement, channel capability, ecosystem standards and certification.
Short observations from live engagements across the region.
We write to be useful to somebody who has to make a decision. We do not write to be quoted in a press release, and we will disagree with a prevailing position when we think it is wrong.
Four editorial rules. They are why this page reads like a practitioner wrote it rather than a content team.
Every piece comes from something we built, measured or had to fix. No thought leadership written by people who have not run an engagement.
Each one is written for a specific role making a specific decision. If it does not help anyone decide anything, we do not publish it.
Definition-first, question-form headings, comparison tables and a stated position — so it can be found and cited, not just read.
Articles carry dates because the field moves. Anything that has been materially wrong gets corrected visibly rather than quietly.
The articles above, condensed into what matters for a decision this quarter. No campaigns, no third-party sharing.
Every topic below is covered in the sections above, and each links to the page that goes deeper on it.
Pick the process you most want to stop running by hand. One working session, one scored diagnostic, and you leave with a decision you can defend to finance.