20260518 - Learning AI the Way I Learned NAV
Areopa webinar (May 2026) in which Nabil Bammou, CEO of Azure Services, explains how he adopted AI using the habits he used to learn NAV. His three lessons are a green, orange, red rule for what to share, going deep on one tool for 90 days, and directing AI instead of chatting with it. He demos a Claude skill that builds a Business Central sales invoice layout from a customer PDF.
- Introduction and housekeeping
- The Pedro Pascal moment - learning NAV through observation
- Deep learning and career trajectory
- The pattern that shapes careers
- Three key learnings - the framework
- From privacy-first to all-in with AI
- Green, orange, red zone framework
- The skimming problem and AI journey
- Tool selection and the Rome crossroads moment
- The 90-day deep focus technique
- Working with AI: from doer to director
- The replacement ladder and director role
- AI-generated invoice layout in action
- Marathons over chats - working with AI as a partner
- Q&A - tools and deep learning approaches
- Wrap-up and personal stories
Overview
Nabil compares learning AI to learning NAV by reading all 400 pages of documentation. He describes moving from privacy-first caution to using ChatGPT with real company and project context, then testing Claude Desktop with MCP and using Claude Code (a research preview) for full-stack work.
He presents a framework: pick one tool and go deep for 90 days, treat AI as a co-worker who needs context, and act as a director who sets standards and verifies output. A demo shows a Business Central print layout skill that builds an invoice layout from a PDF mockup, produced in one long conversation, which he calls a marathon. The session ends with Q&A on tools and deep learning approaches.
Key points
- Privacy rule (green, orange, red): share public info freely, decide quickly on internal strategy or client names, and never share bank details, passwords, credentials or NDA material. Your caution costs more than the risk.
- Pick one AI tool and go deep for 90 days. Do not evaluate competitors until you hit a real ceiling. Skills such as prompting and review loops carry over to other tools.
- Use a marathon conversation instead of one-off questions: explain the full task and context, define a good result, and work end to end in one conversation.
- Demo: a Claude skill for Business Central print layouts read a customer’s PDF invoice, measured it to the millimeter, mapped invoice fields automatically and pushed the layout to a QA environment. The user then checked it side by side against the PDF. It took under 12 minutes of AI time and 2 minutes of the user’s time.
- Director mindset (the replacement ladder): AI does the low-level work while you give context, set the standard, review and ship. AI can hallucinate confidently, so always verify.
- Have AI spawn a code review agent to audit its own scripts for cross-platform issues, such as Mac vs Linux grep differences, and for edge cases.
- Claude Desktop with MCP connects to tools like an Obsidian vault and Airtable. Claude Code came out as a research preview in early 2025.
Features
| Feature | Status | At | Evidence |
|---|---|---|---|
| AI as development partner | status not stated, demoed | 11:29 | |
| Deep learning methodology for complex systems | status not stated | 7:13 | |
| ChatGPT | status not stated | 17:10 | |
| Claude Desktop | status not stated | 22:17 | |
| Claude Code | preview, demoed | 22:33 | “Claude Code came out as a research preview, which took things to a whole new level” (22:33) |
| MCP Protocol | status not stated | 22:17 | |
| Code review agent pattern | status not stated | 27:57 | |
| Business Central print layout skill | status not stated, demoed | 32:06 | |
| Marathon conversation technique | status not stated, demoed | 38:17 | |
| Meta prompting | status not stated | 41:08 | |
| AI director mindset | status not stated | 41:42 | |
| Orchestrator for knowledge management | status not stated | 43:40 | |
| 90-day deep learning methodology | status not stated | 47:32 |
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 “code review agent” at 28:09
- other “Open Claw” at 43:04
- other “Hermes AI” at 43:15
- other “Obsidian” at 43:40
- other “Deep Seek” at 44:44
Quotes
- 5:22 “my friend eventually gives me a tired look and says nabil i have no idea what”
- 7:00 “That moment changed everything. I remember thinking, this is how a pro does it. I need to learn this, and I want to do”
- 7:34 “All 400 pages. The more I read, the more the system made sense, and the more I wanted to go deeper.”
- 12:55 “Even with all the resources I have, I cannot build a private AI that protects my data better than the big labs.”
- 14:10 “I opened ChatGPT and I just start typing. I I’m Nabil. I am the CEO of Azure Services. We help …”
Disclaimers in the video
- 22:33 preview: Claude Code came out as a research preview
- 43:15 other: Hermes AI is described as the new kid in the block, mentioned as something to check out
Presenters (as heard): Nabil, Nabil Bammou, Nabeel, Luke van Vught, Dennis Storøy, Dan, Luke, Claude, Tina.