Blog
Notes on building, evaluating, and deploying KB Sandbox.
Sentient AI?
9/19/2026
Will AI surpass human knowledge? Perhaps in some areas it already has. An AI can process thousands of pages in a fraction of the time it would take a human. It can compare documents, recognize patterns, write software, summarize research and make connections across enormous amounts of information.
The Workbench in Between: Inside the Org, and Outside It
9/9/2026
The thing sitting between a personal bookcase and a filing cabinet: a workbench / workspace.
Shadow AI, From the User's Side
9/9/2026
"Shadow AI" - Good intentions can go wrong.
The Enterprise Shelf: How Organizations Treat Knowledge
9/9/2026
How an organization keeps *its* knowledge
How an AI Model Is Born — and Why Where It Lives Matters
9/9/2026
Foundational Models, fine tuned models, Rag, wikis and agents...where they sit and how they are accessed
Knowledge Is King
9/9/2026
Printing Press - Libraries - Internet - AI knowledgebases
Series Intro: Knowledge Is King
9/9/2026
This is a five-part series about knowledge as the resource behind every consequential era that enabled the renaissance, industrial revolution, digital/internet age and now powering the AI era.enaissance, industrial revolution, digital age and now powering the AI era.
Data Sovereignty Is Not a Setting. It Is a Set of Guarantees You Can Prove.
9/6/2026
A single AI interaction can send your data around the world before you see a response. Data sovereignty is not a reassuring sentence about intent -- it's a set of guarantees you can verify, including whether admin tooling actually respects a restriction too.
Shadow AI Is Already Inside the Enterprise. The Answer May Not Be to Ban It.
9/6/2026
How a governed AI Workbench can give employees the freedom to use AI while giving organizations control over where their knowledge goes.
Practical AI Architecture Governance -- and Where KB Sandbox Fits
8/26/2026
A new piece on proportionate, evidence-based AI governance -- and a look at how KB Sandbox's own Milestone 7 governance pilot could put those same ideas into practice.
Why AI Responses Need Structure, Not Just Markdown
8/26/2026
Separating conversation, summaries, next actions, citations, documents, and trusted links.
Conversation History Is Not the Same as AI Memory
8/26/2026
Designing short-term context, conversation continuity, and long-term memory without confusing users.
Your AI Assistant Should Tell You Which Model Answered
8/26/2026
Why model identity, message-level provenance, and transparent switching matter in enterprise AI.
A Private AI Journal Should Belong to the Individual
8/26/2026
Using conversation history to support reflection without turning personal memory into employer surveillance.
One Default AI Model Is Not Enough
8/26/2026
Conversational reasoning, structured output, embeddings, and evaluation are different jobs.
Private AI Deployment Is an Operating Model, Not a Docker File
8/26/2026
What organisations take responsibility for when they move AI, data, authentication, and storage into their own cloud.
The Document-First Principle for Enterprise AI
8/26/2026
Why an AI assistant should prepare evidence, plans, and reviews before it changes operational systems.
Before the AI-Native SDLC: Test What Is Worth Building
8/26/2026
Faster AI-assisted coding does not make an untested assumption more reliable -- it can just help an organisation build the wrong thing sooner. An independent response to Anthropic's AI-Native SDLC Playbook on the experimental layer that should sit upstream of it.
NotebookLM vs. kbSandbox: From Shared Notebooks to Governed Organizational Knowledge
8/26/2026
NotebookLM excels at source-grounded research. KB Sandbox addresses a different enterprise challenge: curating, protecting, evaluating and operationalizing organizational knowledge.
KB Sandbox Deployment and Server Options
8/21/2026
From Vercel and managed Supabase to Azure, AWS, or a fully private deployment -- how KB Sandbox can move beyond its current hosting without slowing product development.