From the Desks
The latest, by surface.
- 001June 14, 2026Knowledge Hub
The Three Graphs Behind an Agent-First Enterprise
An agent that reaches the edge of what it knows does not stop. It returns a confident, well-formatted, wrong answer, and nobody notices for weeks. The fix is architectural: a semantic graph that binds the shape of your business to the system, table and column already recording it, a process graph that holds the real routing and escalation, and an activity graph that accrues from operations. You author two of them. The third fills itself in. Your data never moves. Here is what each one buys you, the order to build them in, and what you get for free.
- 002June 8, 2026Knowledge Hub
The Compliance Gate: Why AI Agents Have Already Broken the Review Cycle — and What Replaces It
Enterprise AI adoption is outpacing governance at every level. Most major enterprises have already rolled out AI coding agent licences, and those that haven't have shadow AI filling the gap. Simultaneously, regulators across jurisdictions are enforcing the EU AI Act, NIS2, and ISO 42001 with a speed and precision the compliance industry has not seen before. The result is a widening gap between how fast AI is being built and deployed, and how well organisations can demonstrate they are in control of it. Your compliance model is already broken. The question is whether your board knows it yet.
- 003June 7, 2026Knowledge Hub
AI Is Already Writing Your Code. Who’s Governing It?
Your developers are already using AI to write code. The governance infrastructure your organization built for human-authored software was not designed for what is happening now — and a policy document is not a substitute for technical controls. This piece introduces the AI Development Governance Framework: four levels that give every board, CIO, and CTO a precise answer to where their organization stands today, what it costs to stay there, and what it takes to move.
- 004June 3, 2026Knowledge Hub
Your Team Is Ready to Go Live. Are You Asking the Right Questions?
Every AI agent go-live conversation asks the wrong question. 'Does it work?' is a question about the past. The decision in front of a business leader is about what the agent does when the world it was built for no longer exists — and no benchmark answers that.
- 005June 1, 2026Product Reviews
Oracle AI Database 26ai GraphRAG: A Document Graph, Not an Enterprise Ontology
GraphRAG is now a board-level word, and Oracle's version arrives with unusual pull: for many enterprises the data already sits in Oracle, so a few features you already own can turn your documents into a relationship-aware, citable knowledge base with no new platform and no data leaving the building. The proposition is real and, for an Oracle-centric shop chasing trustworthy answers from its documents, it is one of the fastest low-regret moves available. But the word “graph” is doing two different jobs. Oracle 26ai builds a document graph — entities and relationships an LLM extracts from whatever corpus you fed it, shaped by that corpus, excellent for retrieval. It is not a governed enterprise ontology: no shared definition of “Customer,” no reasoning, no constraints, no safe ground for autonomous cross-system agents. The document graph is a property of your documents; an ontology is a property of your enterprise. Buy the first believing it is the second and it fails — not in the demo, but on the third cross-system agent and the first regulatory question it cannot ground.
- 006May 30, 2026Product Reviews
Enterprise Intelligence Platforms — Market Review 2026
The question that decides an enterprise intelligence platform is not which has the best analytics — features change every release — but whether it can serve as a shared foundation every tool and agent connects to, or whether it only governs meaning inside its own walls. Judged that way, on four criteria (semantic scope, external connectivity, access-control architecture, and per-tenant configurability), none of the seven platforms here scores well on all four, and that gap is the finding, not an oversight. Two categories emerge. The ontology-native platforms — Palantir Foundry, Stardog, Graphwise — model typed entity relationships an agent can traverse and, in Stardog's case, infer new facts. The analytics platforms with embedded semantic layers — AtScale, dbt, Cube, Looker — deliver real, immediate metric consistency but govern calculations, not relationships, and each adds another private definition as you add tools. The unmet need across both: no platform ships a pre-built industry ontology you can configure and go live on in weeks, which is where the largest value, for buyers and vendors alike, still sits.
- 007May 19, 2026Knowledge Hub
LLM-Powered Knowledge Graphs for Enterprise Intelligence and Analytics
An enterprise's most valuable knowledge — the decisions made in meetings, the expertise shown in email threads, the commitments buried in chat — is also its least searchable, and it walks out the door when people leave. Vector search alone cannot recover it: ask “who is the expert on Project X?” and you get documents that mention Project X, not a ranked list of people. This research builds an activity-centric knowledge graph from exactly those internal sources, using an LLM to extract entities and infer the cross-silo relationships that were always there but hidden, then uses retrieval as the natural-language way in. The combination answers relational questions neither method can answer alone. Two cautions decide whether it works in practice: LLM relationship inference gets noisy on messy real-world communications, and a unified graph of everything the company says is among the most sensitive artefacts it can build — access governance is a prerequisite, not a later detail.
- 008May 1, 2026Knowledge Hub
Proxy-Pointer RAG: Structure Meets Scale at 100% Accuracy with Smarter Retrieval
Most enterprise RAG underperforms on the documents that matter most — contracts, filings, manuals, compliance reports — for one reason: shredding a document into flat chunks throws away where each chunk lived, and the heading hierarchy of a 10-K (Company > Financial Statements > Cash Flows) is itself an implicit knowledge graph the author already built. Proxy-Pointer stops destroying it. It parses headings into a skeleton tree, prepends each chunk's full structural breadcrumb before embedding, then retrieves in two stages — broad semantic recall, followed by an LLM re-ranker that scores candidates on structural relevance rather than surface similarity — and feeds the synthesiser complete sections instead of truncated passages. It needs no GPU and no graph database; it slots into an existing vector stack, and every answer cites the section path it came from, which is an audit property in regulated work. Two honest caveats temper the headline result: the perfect score is on a small, author-built benchmark, and the whole approach is upper-bounded by heading quality — it shines on well-structured documents and falls back to ordinary RAG on scanned or flat ones.
- 009April 10, 2026Knowledge Hub
Knowledge Graph vs Semantic Layer vs SQL/PGQ: The 2026 Enterprise Architecture Guide for AI Agents
Three different things are sold under one name. A full materialised knowledge graph, a semantic layer with a virtual ABox over your existing systems, and a thin SQL/PGQ wrapper that makes a relational database behave like a graph are not interchangeable — they differ wildly in cost, lock-in, and time-to-value, and vendors blur the line on purpose. Agents reset the stakes: a thin wrapper alone gives an agent structure without meaning, and an agent acting on the wrong definition of “Customer” is a remediation incident, not a meeting-room debate. So the semantic layer becomes the minimum grounding layer for any agent of consequence, a full graph earns its cost only where relationships are the product or agent memory needs structure, and the real deliverable is a deliberate hybrid. The discipline that decides success is unglamorous: don't buy the ontology before you hire the ontologist, and don't ship the agent before you buy the ontology.
- 010January 30, 2026Knowledge Hub
The Case for a Central Enterprise Ontology
Most enterprise AI is succeeding at the task level — one agent, one job, useful output — and failing at the seam between tasks. When agents from different systems share context or hand off work, the absence of a shared foundation turns visible and expensive: “revenue” means one thing in Power BI and another in Salesforce, one agent's “customer” is a billing account and the next reads it as a CRM contact, and the wrong downstream action follows. The fix is a central enterprise ontology — a formal, queryable model of what the business is, that every tool and agent connects to instead of redefining. Its cost is paid once; its value is collected on every agent deployed after, reversing the usual trajectory where each new system deepens the inconsistency. The discipline is knowing what not to build: the regulatory and industry layers are shared infrastructure, not your IP, so only the enterprise-specific layer is yours — and the place to start is the codebase, where the business is already precisely modelled and the agents creating the risk already run.