AI basics: what the technology actually does
Understand AI, machine learning and generative AI before choosing a business tool.
Business Central
Claude Code and premium models in Business Central development
Separate the coding tool, the model and the deployed extension—and keep engineering checks in the loop.
Independent explanation by Frontier AI Works. Product guidance checked 12 September 2026. Business examples are illustrative.
Claude Code is a development tool
Claude Code is Anthropic's agentic coding environment. With the access you allow, it can inspect a repository, propose and make edits, and run development commands. In Business Central work, useful tasks include explaining existing AL code, drafting a bounded change, suggesting test cases and reviewing an integration. It does not replace the AL toolchain, Business Central permissions or a developer's responsibility for the resulting behaviour.
The tool and the model are different choices
Claude Fable 5.1 is an Anthropic model; GPT-6 Astra is an OpenAI model. They are not additional Microsoft agent-building portals. Claude Code and an OpenAI coding environment are tools through which supported models can perform development work. Do not assume Astra is a native model option inside Claude Code, or that a model available from its provider is automatically available in a particular Foundry region or tenant. Check the chosen product's current model support, access conditions and billing before designing around it.
Two meanings of an AI extension
An extension written with AI assistance is not necessarily an extension that calls AI when users run it. The first may contain ordinary AL code with no model dependency at runtime. The second needs an explicit service integration, authentication, data-handling decisions, error behaviour and ongoing usage costs. A Claude Code subscription used by a developer does not automatically license or host AI calls for Business Central users.
A useful development sequence
Begin with a small requirement and acceptance criteria. Supply the assistant with the target Business Central version, relevant AL objects, extension dependencies and coding conventions. Review its proposed changes, compile with the supported AL tooling, run relevant tests and validate in a sandbox before release. For an illustrative EDI change, test duplicate messages, missing identifiers, unit conversions and retries—not just one successful sample. Generated code can compile while still implementing the wrong business rule.
Choose capability by evidence
A premium model may be worth evaluating for unfamiliar code, difficult debugging or changes spanning several components. Compare models on the same bounded task using correctness, review effort, elapsed time and total cost. A stronger model does not remove the need for tests or approval before deployment. Keep secrets out of prompts and repository instructions, and provide only data the organisation has approved for the chosen service. Our approach is to combine AI assistance with Business Central engineering discipline, rather than promise an unmeasured productivity percentage.
Microsoft sources and further reading
Put it into practice
Business Central Development with AI
Bring AI assistance into planning, coding, testing and documenting Business Central extension changes.
Explore the programme ↗