Episode 515: Stop Coding, Start Architecting: How AI Is Reshaping the BC Developer Role
Dynamics Corner episode 515 with Vladimir, a Ukrainian BC developer and Microsoft MVP, on how AI changes AL development. Topics include AI-generated code and tests, TDD, planning mode, context and session management, AL-Go CI/CD, MCP, and choosing between Copilot, Cursor and command-line agents. It also covers the 2026 wave one agent preview, the troubleshooting MCP server for AL (due in May) and his open source tools Data Editor and Parallel Task Worker.
- Introduction and guest welcome
- 2026 wave one features and agent preview
- Guest introduction and background
- Developer tools and open source projects
- Contributing to Business Central
- AI impact on AL development and code generation
- Risks of generating too much code and quality control
- AI-generated tests and TDD for AL development
- Test generation on production projects
- Testing value and client communication
- Quality improvement and testing across extensions
- AI-assisted development challenges, planning mode, and context management
- Tool selection, benchmarking, and team workflow management
- CI/CD and AL-Go for multi-developer projects
- AL-Go benefits and getting started with AI tools and MCP
- IDE and command-line agents: Cursor, VS Code, and practical MCP approach
- AI development landscape, architecture knowledge, and BC developer skills
Overview
The guest, a Business Central developer from Ukraine and recent Microsoft MVP, discusses how AI makes code cheaper to produce and shifts developer time toward architecture and understanding Business Central functionality. The episode opens with 2026 wave one items, including the agent preview and agent playground, and a troubleshooting MCP server for AL that is due in May. It then covers open source tools such as Data Editor and Parallel Task Worker.
Most of the discussion is about working practice. It covers AI-generated tests and TDD for AL, planning mode before generating code, context and session management, and AL-Go for CI/CD with several developers. It also compares IDE agents and command-line agents, and advises trying MCP in practice. The Business Central MCP is not recommended for production, only for sandboxes or Docker environments.
Key points
- Agent building in sandboxes is in preview. Generating code from agents built visually is not yet automatic.
- The troubleshooting MCP server for AL is due in May 2026. Business Central MCP is for sandboxes or Docker only, not production.
- Generated code still has to be read and understood. Generating too much code without review causes quality problems.
- Tests are now standard for AL. AI can generate and suggest tests, and TDD is recommended, but tests must be reviewed and must fail and pass correctly.
- Start with planning mode: give the AI full context, name the relevant objects and say why they matter. Use separate sessions for different tasks.
- AL-Go is a free CI/CD tool that runs in GitHub. GitHub runners may cost money depending on setup, and branch discipline is required.
- Tool rankings change often. The BC bench from Microsoft counted only two agents, so evaluate tools yourself. Source control and code review matter more than which tool was used.
Features
| Feature | Status | At | Evidence |
|---|---|---|---|
| Agent preview and agent playground | preview, demoed | 3:59 | “Maybe we’re still in preview. That’s what they want though. That’s what they want, right?” (6:42) |
| Unified interfaces for AI | status not stated | 4:26 | |
| Enable troubleshooting MCP server for AL | status not stated | 5:03 | |
| AL tasks from Visual Studio Code | status not stated | 5:16 | |
| Data Editor tool | status not stated | 9:23 | |
| Parallel Task Worker | generally available | 9:53 | “recently I released um parallel task worker. It’s like a library for developers uh where you can able to uh easily to split some” (9:53) |
| AI-assisted code generation for AL | status not stated | 14:27 | |
| Business Central MCP | status not stated | 17:31 | |
| AI-generated test creation for AL | status not stated, demoed | 17:47 | |
| TDD approach for AL with AI | status not stated, demoed | 18:38 | |
| AI-generated test suggestions and creation | status not stated, demoed | 20:24 | |
| Automated testing for production code safety | status not stated, demoed | 22:51 | |
| AI-assisted documentation creation | status not stated | 25:24 | |
| Quality improvement through AI-assisted CICD and best practices | status not stated | 26:59 | |
| Testing for Internal and External Extensions | status not stated | 28:00 | |
| AI Challenges in Collaborative Development | status not stated | 29:57 | |
| AI-Assisted Planning Mode | status not stated | 32:36 | |
| Context Window and Session Management | status not stated | 33:31 | |
| AL Development Benchmarking | generally available | 35:49 | “Microsoft released a BC bench for agents. And they it was as I remember they only count two agents, Cloud Code and maybe Codex” (35:49) |
| Source Control and Code Review Workflow | status not stated | 40:28 | |
| AL-Go free tool for CI/CD | status not stated | 42:15 | |
| MCP protocol for AI communication | status not stated | 47:02 | |
| Copilot in Visual Studio Code | status not stated | 48:41 | |
| Command-line AI agents | status not stated | 49:06 | |
| Cursor IDE | status not stated | 48:12 | |
| MCP in Practice | status not stated | 52:30 | |
| AI-Assisted Development with Architecture Review | status not stated | 53:13 |
AL objects mentioned
As heard in the captions. A name that matches one object page by exact type and name links to it; the others stay as named.
- codeunit “unified interfaces code units for AI” at 4:26
- codeunit “library code unit” at 22:21
- other “BC bench” at 35:49
- other “Codex” at 36:05
- other “Cloud Code” at 36:05
- other “Cursor” at 36:42
- other “Copilot” at 36:42
- other “MCP” at 47:02
- other “AL-Go” at 42:40
- other “Business Central MCP” at 51:19
Not found in BC28-30: codeunit “unified interfaces code units for AI”, codeunit “library code unit”.
Quotes
- 3:59 “now you can build some agents in your sandboxes environments”
- 4:36 “Some of Some of my past experience was great, some of it’s not great. Like there’s some limitations like you cannot pass files and”
- 5:16 “I know that’s due in May, so I’m I’m looking forward to seeing that”
- 5:41 “I really do hope that they make it to where you can create the agents and not have to create the code for the”
- 7:10 “I’m Vladimir. Uh ABC developer from Ukraine, a recent Microsoft MVP. Uh and I have been in the ecosystem for 9 years”
Disclaimers in the video
- 3:59 preview: now you can build some agents in your sandboxes environments
- 4:36 subject-to-change: it’s still in in the stage where it’s evolving. And there’s some challenges which we are uh right now trying to to solve
- 5:16 coming-later: I know that’s due in May, so I’m I’m looking forward to seeing that
- 6:42 preview: Maybe give it a a few months. Yeah, it just it just very fresh techno Yeah. Business Central very fresh. …
- 51:19 not-in-this-release: Business Central MCP not for production use. Just do your daily work on some sandboxes or on some uh Docker environment.
Presenters (as heard): Chris, Brad, Vladimir, Al.