I Need Coffee - Episode 218 - Weekly BC Review! (July 24, 2026)

1. App & Product Work
   - Corporate Direct Debit BAI2 import marketplace app: on-the-fly enhancements after customer clarified real requirements (A *and* B, not just A). BAI2 parsing itself is now easy with AI; process/edge cases are the hard part.
   - New BAI2 user guide auto-generated, versioned (v005), committed to repo, and published nightly by the documentation site.
   - Business Central ↔ bill.com integration (with MEM support): new feature added via Fable/Claude to work around a bill.com limitation. Full docs and user guide updated.
   - Customer PTEs reviewed with partner: complex line-level sales price lists (customer price group, multiple levels — not standard BC lowest-price logic) and BOM explosion that retains parent-child relationships, original assembly price, and quantity cascading. Feature-complete and well received.
   - bill.com marketplace app sandbox cleanup: scrubbing customer data for a usable demo/test environment (hard with most SaaS platforms).

2. DevOps & Infrastructure
   - Switching from BC Container Helper / Algo-for-GitHub (painful artifact/image downloads, 30–60+ min pipelines, >2,000 GitHub Actions minutes burned in <2 weeks on one app) to Cosmo Pack / Cosmo Alpaca (target ~5 min runs). Now enabled on GitHub Enterprise orgs.
   - Code-signing certificate move: DigiCert reissue triggered organization re-verification (messy process, outdated Skype references, wrong product type in emails, high cost ~$700). Evaluating Azure Artifact Signing ($10/month, available in US/Canada/EU/UK, GitHub + Algo integration, automated identity verification via Microsoft Authenticator + third-party ID service with face/driver’s license scan).
   - New ERP Center of Excellence website (site #4): rebuilt with Claude Code CLI + Astro 6 + CSS + GitHub Actions → Cloudflare Pages. Slideshow, more photos/content (still trimming), sub-200 ms loads, auto-publish in ~60 seconds. Original 2023 Namecheap free builder site retired.

3. AI Tooling Observations & Quality
   - Fable/Claude highly effective for complex feature work + docs once prompted properly; Anthropic limit resets and team credits helped usage.
   - Claude often produces dense technobabble / “AI meta language.” Explicit prompts for plain English, no jargon/metaphors/quips dramatically improve results (occasional residual “Honest answer, Steve…” style slips remain).
   - Marketing research paper example: first AI draft was abstract-heavy and audience-mismatched; rewritten in simple language for a new marketing associate / ESL reader.
   - Persistent missing feature across AI tools: no timestamps on messages/prompts/responses. Host implemented a workaround via Claude settings.json hooks (user-prompt-submit + stop) + PowerShell, forcing the LLM to emit timestamps. Works better in CLI than VS Code extension; still a duct-tape solution.
   - Quality-noticing exercises: Joshua Bell Stradivarius subway video (most people walk past excellence); Blue Dragonfly logo deliberately distorted ±15% aspect ratio — once seen, similar squash/stretch errors appear everywhere on signs, menus, vehicle wraps, etc.

4. Bugs & Technical Discoveries
   - Obscure latent bug (Franklin): removing a permission set causes one specific event subscriber to silently fail while everything else works. Investigating registration library vs. permission-set interaction.
   - Learned about Permission Set Extension objects (BC 2021 Wave 1). Caution advised for extending base/ISV sets; safer for own library apps.
   - Microsoft Teams Planner task chat: appears broken (messages not visible until refresh / one-way visibility). Described as “duct tape all the way down” on an eventually-consistent distributed service that has been passed around Microsoft orgs.

5. Cybersecurity Notes
   - Nextcloud site compromised via WordPress vulnerability (public PSA: WordPress + plugins have had >2,000 vulnerabilities in recent years; treat as high-risk for anything important). Host recommends static sites (Astro → Cloudflare) instead.
   - Researcher “fried rice” / Chaoan / Shiao Shu (who previously found Claude source code in an npm package) is using Kimmy K3 (open-weight, minimal guardrails) to discover zero-days rapidly (e.g., Redis, Telegram Desktop/iOS one-step-from-RCE, calculator pop in minutes with multi-agent setups).
   - GitHub paid a $100k bounty (via HackerOne) for an RCE on github.com.

6. Team Updates
   - Joselyn: continuing financials / GL data-load config packages; sat MB-800 again (failed on purchasing/inventory sneaky questions); studying fixed assets, posting groups, deferrals.
   - David: check report work + continued studying.
   - Ronald: heavy lift on Berlin ERP Academy curriculum (filling gaps in AI-generated skeleton materials — adding visuals, screenshots, videos, task-level assignments for LMS); also running second round of Berlin English Academy for E4CC English scholarship graduates (practice speaking).
   - Franklin: permission-set bug research; starting a relatively small PTE tomorrow; user-guide updates.
   - Fatima: supporting Berlin ERP Academy; finalizing next El Salvador Doug meetup (around Aug 19); Teams Planner exploration.
   - Daniela: marketing pivot from traditional campaign/metrics approach to gradual brand-building; aiming for first campaign next week. Office visitor: a mouse crawled on her during a call.
   - Overall office wildlife in Berlin, El Salvador includes cats, dogs, birds, mice, etc.

7. Upcoming Events & Personal
   - Vive 2.0 — Provo, Utah (September); host speaking (two sessions). ~45–60 min south of Salt Lake City.
   - Summit / conference — Nashville, October 11; book hotel/flights/register.
   - Dynamics Con regionals: Ghana (July 25 — check remaining availability), Columbus (August 25 — double-check dates).
   - Personal: early Friday ~12-hour / 700+ mile drive to Utah (Heber / Park City area) for house maintenance + time with Tanya Henderson. Forecast high 90s °F at elevation; scenic drive past Jordanelle Reservoir.