Business Central Musings

I’ve been writing C/AL and AL code for years. Table extensions, codeunits, page customizations — I could scaffold a Business Central extension in my sleep. I’ve published over 100 blog posts breaking down BC development patterns, recorded YouTube walkthroughs, written how-to guides and exam cram booklets for the MB-820. I’ve invested thousands of hours building and sharing AL knowledge. A few weeks ago, I asked Claude Code to do the same thing. It took 90 seconds. And the code was correct.

That stung.

With Opus 4.5, GPT-5.2, and Gemini 3 all crossing a capability threshold in early 2026, the reality is brutal and simple: writing AL code is no longer the scarce skill. The models write subscribers, event publishers, and upgrade codeunits faster than we do. They know the BC base app patterns. They read Microsoft’s docs better than most of us ever will.

So what’s left for us?

Everything that matters.

The models can write AL. They cannot decide what to build or why. They cannot stare at a client’s broken warehouse process and realize the real problem is the approval workflow, not the posting routine. That’s product thinking — and it’s now your primary job.

Here’s what I believe the new AL engineer looks like:

1. Product-minded, not syntax-minded. You understand the business domain deeply enough to define what the extension should do. The AI writes it. You validate it. The skill is judgement, not keystrokes.

2. Test-driven by default. If AI writes your code, how do you trust it? Test Codeunits and BCPT are no longer optional. They’re your safety net. Write the test first, let the agent write the implementation, run the suite. That’s the loop now for now.

3. Fluent in Azure. Application Insights for telemetry. Logic Apps for integration orchestration. Azure Storage and Data Fabric for data pipelines. The BC extension is just one node in a larger architecture. If you only know AL, you only know the leaf.

4. Building agents, not just extensions. The developers pulling ahead are orchestrating full product lifecycles with AI agents — from spec refinement to code generation to test execution to deployment. Claude Code, GitHub Copilot Workspace, and similar tools aren’t assistants. They’re team members you manage.

The skill of hand-writing AL code came with a deep satisfaction. But the developers who cling to “I write the code” as their identity will be outpaced by those who embrace “I am responsible for the outcome.”

The AL language isn’t going anywhere. Your role in it already has.

Buy the book: Human, Upgraded: Stop Panicking, Start Sprinting