Resources · Articles
Articles
Plain explanations of the ideas behind enterprise AI: what each one means, how it works, and where it matters, with the sources behind them.
AI factories
What is an AI factory?
The two meanings of the term, the six stations a request passes through, and what a factory needs underneath before it ships anything.
AI factories 02Why should a person approve what an AI changes?
Human in the loop and on the loop, where approval belongs and what a reviewer must see. And what the EU AI Act and NIST ask for.
Human oversight 03What is AI governance for internal tools?
Who may build, review, ship and change the tools AI builds inside a company, the five controls that answer for them, and where the EU AI Act fits.
AI governanceGraphs and context
What is a knowledge graph?
Entities, typed relationships and a schema. How a knowledge graph differs from a table and a search index, and how to tell a useful one from a diagram.
Knowledge graphs 05What is a context graph?
Two meanings of a new term, decision traces and a permission-aware graph of records, where they meet, and where a knowledge graph ends.
Context graphs 06What is context engineering?
Choosing what goes into a model’s context window at each step and what stays out, and why a bigger window does not help.
Context engineering 07What is entity resolution?
Matching keys, fuzzy matching and the two cutoffs between them. Why a false merge costs more than a missed match, and why a graph needs it.
Entity resolutionSearch and knowledge
What is enterprise search?
How search inside a company differs from web search, how it works, where it goes wrong, and how to test one before you trust it.
Enterprise search 09What is hybrid search?
Keyword and meaning-based search on one query, how their two lists become one ranking, and when an exact match has to win.
Hybrid search 10What does a search score mean?
Why the number next to a search result orders a list but grades nothing, how it is built, and why two searches’ scores do not compare.
Search scores 11What is AI knowledge management?
How AI changes capturing, organizing and finding what a company knows, how it differs from a knowledge base and from search, and what goes wrong.
Knowledge management 12What is intelligent document processing?
How PDFs, scans, Office files and email become text, structure and fields: OCR, layout, classification, field capture, and where a person still checks.
Document processingAgents
What is an AI agent?
The loop an agent runs, how it differs from a workflow, and what an enterprise agent needs before it touches real records.
AI agents 14What is an agent sandbox?
Why an agent that runs code needs walls, the five common ways to build them, and what a sandbox has to limit to be worth the name.
Agent sandboxes 15What is the Model Context Protocol?
Hosts, clients and servers, the tools, resources and prompts a server offers, and what the protocol leaves to you: who may connect and what a model may do.
Model Context Protocol 16MCP, AI tools and agents: what is the difference?
A tool is one function, an agent is a loop that calls tools, and MCP is a standard way to offer them. How the three stack, and when to build which.
Tools, agents and MCP 17What are agent skills?
A folder of instructions an agent reads only when a task needs it. The SKILL.md format, progressive disclosure, and how a skill differs from a tool or a system prompt.
Agent skills 18What does a coding agent need from a platform?
A coding agent writes the app. The seven things it has to stand on before the app ships, how each goes wrong, and how to check a platform for them.
Coding agentsModels and infrastructure
What is a multi-tenant AI platform?
Many customers or teams on one platform. The three isolation models, what leaks when isolation fails, noisy neighbors, and the questions to ask a vendor.
Multi-tenancy 20What is a model agnostic AI platform?
Why the model should be a setting, how a provider layer makes it one, and what still breaks when you change it.
Model choice 21What is AI vendor lock-in?
The five layers an AI platform collects your work in, what leaving costs at each, and how to test the exit before you sign.
Vendor lock-in 22How do you serve open weight models on your own hardware?
The model file, quantization, memory as the limit, the API your code calls, and one request at a time against batching.
Open weight models 23What is air gapped AI?
AI on a network with no connection out. How it differs from on premises and private cloud, what has to come inside, and how files cross the gap.
Air gapped AISecurity and sovereignty
What is permission-aware retrieval?
Why an assistant may only read what the person asking may open, where the access check belongs, and how permissions go stale.
Permissions 25What is relationship-based access control?
ReBAC against RBAC and ABAC, what Google’s Zanzibar paper showed, permissions as edges in a graph, and why the check runs when the question is asked.
Access control 26What is prompt injection?
Direct and indirect injection, why no filter stops it reliably, the lethal trifecta, and the designs that limit what a steered assistant can do.
Prompt injection 27What is sovereign AI?
The five layers of control, the national and the enterprise meaning, and what EU and US law say. Then a test for any deployment.
Sovereign AI 28What does sovereign AI mean in Europe?
The EU rules that give sovereignty its content: GDPR transfers, the reach of the US CLOUD Act, the AI Act and the Data Act. Then what the EU is building, and five questions for a platform.
Sovereign AI in EuropeYour data. Your infrastructure.