Episode 521: 50 Developers, Zero Playbook: The Leadership Conversation Nobody's Having
Podcast discussion on leading large Business Central developer teams through fast AI change: conference content gaps, standards and governance, and managing AI tool costs. Speakers describe multi-model use, GitHub Copilot's move to credit-based billing, and assigning models by developer experience.
- Introduction and seasonal reflections
- Pace of technological change and conference timing challenges
- Building sessions that stand the test of time
- Guest introductions: Matt Traxinger and Tanya Brockmann
- Leadership and development session gap at conferences
- Team governance and managing change with multiple developers
- Conference types, audience segmentation, and content strategy
- Balancing partner vs end-user developer focus at conferences
- Rethinking conference track categories and AI blending into administration
- From academic to functional conference content and AI as dominant topic
- AI adoption strategy and shifting from technical deep-dives to real-world application
- Panel sessions and open forums as preferred conference format
- Managing large teams during rapid AI feature adoption and framework for AI adoption and GitHub billing changes
- Improving existing solutions versus chasing new projects
- Using AI for testing and edge cases instead of speed
- Measuring quality and delivering documentation with AI
- AI usage patterns beyond development coding
- Multi-model approach, team adoption strategies, and tokenization costs
- GitHub Copilot pricing model and business considerations
- Managing AI credit costs and team education
- Responsible AI usage and credit allocation strategy
- Best practices for selective AI tool deployment
- Model selection strategy for cost optimization
- AI model differences, training data concerns, and LLM experimentation
- Learning from base code and AI-generated content
- Closing remarks, contact information, and episode credits
Overview
Episode 521 of the Dynamics Corner podcast covers two connected themes. The first is conference content: the guests say sessions that sit between leadership and pure development, such as team adoption, standards enforcement and change management, tend to be rejected, and they favor panels and open forums over lectures. They also discuss how AI is blending administration and development, and how the pace of change makes it hard to plan sessions months ahead.
The second theme is running a team of about 50 developers while AI features arrive quickly. Speakers describe using AI more for design, testing and documentation than for raw coding speed, combining several models for design and review, and the cost pressure from GitHub’s switch to paid credits. They discuss giving costly models to experienced developers and cheaper ones to others, and raise concerns about AI models training on AI-generated content.
Key points
- A speaker’s team has about 50 developers and lets customers write code alongside them. Getting internal and client developers to follow the same standards is hard, especially as AI features ship quickly.
- Sessions that fit neither development nor leadership, such as team adoption, standards and change management, are said to be rejected more often at conferences.
- Speakers prefer panels, open forums and Q&A over lecture-style sessions, and want conference content to be more functional and less academic.
- Matt says he uses AI for everything and for development the least. He spends more time on solution design, and coding ends up as about 20% of total time.
- One speaker does initial design with Claude, uses ChatGPT as design reviewer and then Gemini for a further review, so the layers converge on a solution.
- Copilot or Claude Code is described as getting about 80% of the way there. The developer still fixes Business Central-specific details such as wrongly named fields and posting setup references.
- GitHub Copilot moved from a monthly plan with a fixed allowance to actual credit use, which makes cost harder to estimate. Tanya notes that splitting a large codeunit into smaller ones uses fewer tokens.
Features
| Feature | Status | At |
|---|---|---|
| AI agents for administration | status not stated | 18:57 |
| Sales order agent | status not stated, demoed | 20:12 |
| Telemetry component in administration | status not stated | 17:41 |
| Custom agents for Business Central | status not stated | 28:18 |
| AI-assisted code unit optimization | status not stated | 28:43 |
| AI for Business Central performance testing | status not stated | 32:11 |
| AI-generated documentation | status not stated | 34:15 |
| AI for solution design and architecture | status not stated | 36:21 |
| Multi-model AI approach for design review | status not stated | 38:52 |
| Copilot for code generation | status not stated | 39:59 |
| GitHub Copilot credit-based pricing model | status not stated | 42:25 |
| Responsible AI model usage by developer experience level | status not stated | 47:22 |
| Code review agent | status not stated | 48:50 |
| Cheaper AI model usage for documentation and code reviews | status not stated | 51:09 |
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.
- other “GitHub Copilot” at 3:31
- other “Visual Studio Code” at 3:31
- codeunit “event subscribers codeunit” at 28:43
- other “Business Central Performance Toolkit” at 32:27
Not found in BC28-30: codeunit “event subscribers codeunit”.
Quotes
- 3:20 “there’s so much coming out that it feels like it’s changing by the minute because I can’t keep up with the volume.”
- 4:36 “It is a challenge coming up with sessions so far in advance because you want to talk about things that are relevant at the”
- 8:38 “Over the years, I’ve noticed that the ones that don’t quite fit development and don’t quite fit leadership get rejected more often than than”
- 11:50 “we have like about 50 developers. We allow our customers to, you know, write their own code and contribute to their apps alongside of”
- 12:42 “it’s very grounded in that content and it’s like definitive like you can say this”
Disclaimers in the video
- 29:36 other: There’s no like standard. You don’t feel like you are really grounded in your standards because your standards are always …
Presenters (as heard): Chris, Brad, Matt, Matt Traxinger, Tanya Brockmann, Tanya, Tony, Matt Traxler.