https://stefanmaron.com/tags/ai/Recent content in AI on Stefan Maron | Business Central & AL DevelopmentHugo -- gohugo.ioMon, 10 Aug 2026 07:00:00 +0200https://stefanmaron.com/posts/bc-code-atlas-grounded-search-for-agents/Mon, 10 Aug 2026 07:00:00 +0200https://stefanmaron.com/posts/bc-code-atlas-grounded-search-for-agents/Ask an AI coding agent where BC validates a sales order before posting, and it will confidently give you a procedure name. Sometimes that name is real. Sometimes it’s a plausible-sounding hallucination assembled from training data that’s a few versions stale. Either way, the agent isn’t looking anything up — it’s pattern-matching against what it remembers, and it has no way to tell you which one it’s doing. I’ve been building bc-code-atlas to fix that, and I’m announcing it publicly today.https://stefanmaron.com/posts/al-runner-run-al-tests-without-bc/Fri, 24 Apr 2026 07:00:00 +0200https://stefanmaron.com/posts/al-runner-run-al-tests-without-bc/One Sunday afternoon in March I was playing with my daughter when a thought struck me out of nowhere: what if we could run AL code directly? Not through the service tier, not through a container — just… run it. I fired up the PC, set Claude Code loose on the investigation, and went back to playing. By the time we were done, there were some very interesting findings waiting for me.https://stefanmaron.com/posts/swappable-claude-profiles-container-mounting-part-3/Mon, 16 Mar 2026 09:00:00 +0100https://stefanmaron.com/posts/swappable-claude-profiles-container-mounting-part-3/After Part 2 we had a working standalone container — no VS Code required, no IPC socket surprises. But there was still one problem: a single global ~/.claude folder shared across everything. One set of instructions, one list of skills, one set of agents — regardless of whether you’re doing AL development, debugging telemetry, or investigating something completely different. That doesn’t scale. So this part introduces profile folders. You can watch the full stream on YouTube if you want to follow along.https://stefanmaron.com/posts/turning-streams-into-blog-posts/Tue, 24 Feb 2026 16:00:00 +0100https://stefanmaron.com/posts/turning-streams-into-blog-posts/I’ll be honest: I’m lazy when it comes to writing blog posts. Not lazy about sharing — I try to stream regularly, push code to GitHub, drop things in the community Discord. But sitting down to write a structured article about something I just spent two hours live coding? That rarely happens. The stream is the content. I jump on, explain what I’m doing, answer questions live, push something to GitHub, maybe add chapters via AI later.https://stefanmaron.com/posts/claude-code-dev-container-al/Fri, 20 Feb 2026 09:00:00 +0100https://stefanmaron.com/posts/claude-code-dev-container-al/A few weeks ago I shared my Claude Code configuration for AL development — the CLAUDE.md instructions, the agent profile, custom commands. It’s all public and you’re welcome to use or adapt it. But getting it into a new project still involves some manual steps, and there’s one thing that always annoyed me: I was running Claude Code outside my dev containers. Not because I wanted to, but because I didn’t want to authenticate inside every new container.https://stefanmaron.com/posts/al-development-claude-code-multi-agent-workflow/Wed, 28 Jan 2026 09:00:00 +0100https://stefanmaron.com/posts/al-development-claude-code-multi-agent-workflow/I’ve been using Claude Code quite a bit over the past year, and I think I finally have a process that actually works for AL development. One-off prompts don’t cut it for anything non-trivial — AL is niche enough that LLMs struggle without guidance, and large refactoring tasks blow up the context window within a few agent calls. What changed things for me was switching to a sub-agent workflow. You can watch the full stream on YouTube if you want to follow along.https://stefanmaron.com/posts/businesscentral-sentinel-documentation-stream/Mon, 09 Dec 2024 09:00:00 +0100https://stefanmaron.com/posts/businesscentral-sentinel-documentation-stream/This is episode two of the BusinessCentral.Sentinel coding stream. If you missed the first one, Sentinel is a free AppSource app that runs a set of checks against your Business Central environment and flags potential issues — extensions in DEV scope, missing source code access, evaluation companies in production, that kind of thing. A lot happened between the two streams. I shipped a release off-stream, fixed a bug where extensions weren’t showing a name in some sandboxes, and got a community pull request.