Blog
Practical writing on Business Central, AI and the work in between.
Each post names what it is grounded in. Sources sit at the foot of the page with the date we last checked them.
Copilot and connections
- Copilot in Business Central and UNI4C Controller: what each one answersCopilot is included in your Business Central licence and does many tasks well. Controller answers questions across your ledger and verifies each figure. This page compares them.7 min read
- The Business Central MCP server: what it does and what you decideMicrosoft's MCP server gives AI clients read-only access to your API pages by default. Three switches change that. This page explains each switch and the risk it carries.6 min read
- Connect ChatGPT or Claude to Business Central: four things to solveA direct connection is supported and it works. Permissions, audit records, truncated reads and unverified figures are the four problems that remain after you connect.4 min read
Grounded answers
- The assistant writes the query, not the numberA language model that holds your figures can change them. This page explains a design that keeps the model away from the values and puts the platform between them and the reader.4 min read
- Five sentences about a ledger, and what survived verificationFive cases from the figure filter's own test suite. Each case shows the sentence the model wrote and the sentence the reader is allowed to see.4 min read
- Truncation is the quiet bug in ERP AIAn assistant that read the first 500 of 3,901 rows can still write a confident total. The answer looks correct. This page explains how to find the fault and how to stop it.3 min read
- When the correct answer is a questionThree customers match the name. The period is not clear. The measure has two readings. A good assistant asks. This page explains when to ask and how to ask well.3 min read
Europe, law and data
- AI Act Article 50 and AI over your ERPArticle 50 applies from 2 August 2026 and is not limited to high-risk systems. This page explains the duties for a company that runs an AI assistant over ERP data.4 min read
- Where your ledger data goes when you ask a questionThe path of one question through the platform. What leaves your tenant, what reaches a model, what the model provider keeps, and what never moves at all.4 min read
- What a retention window deletesA conversation is not only its messages. This page explains what a 14-day window removes, why the queries go with it, and why business terms have their own setting.3 min read
Practice
- Read Business Central telemetry without KQLThe event IDs that carry the story, what each one means, and the Log Analytics limits that surprise new users. A practical guide for consultants and administrators.4 min read
- An AL extension written by an agent: what the review must catchAn agent can write AL and publish it to a sandbox. This page lists what a consultant must check before that code reaches a production environment.5 min read
Foundations
- What is a semantic layer for an ERP assistant?A definition, and the reason the layer must build itself. A semantic layer that needs a project to fill it stays empty, and an assistant without one guesses what your terms mean.5 min read
- Five levels of AI access to an ERPA ladder from a nightly export to a live query that runs with the permissions of the person who asked. Use it to place any vendor, including this one.5 min read
- What AI over Business Central costs for each userSeat prices, the included AI allowance, and the rule that stops a surprise bill. Copilot is included in the Business Central licence, so this is what sits beside it.4 min read