Just had a great conversation that covered the entire arc of software development — from legacy systems to modern AI-assisted shipping.
Brad walked through how he shipped his first live web app running on Windows Server / IIS with a Microsoft SQL Server backend — largely guided by AI prompts, with real-world corrections applied using deep Windows experience.
We then zoomed out into why vibe coding feels like a full-circle moment: after years in executive and business roles, AI agents make it possible to return to building without being blocked by today’s frameworks.
What we covered
- Shipping a live IIS + SQL Server app with AI-assisted deployment
- Career arc: COBOL, punch cards, IBM System/360 → modern web apps
- Why shipping something live (even basic) matters
- Working through Corbin Brown’s Thumbio course
- Using vibe coding to rapidly recreate complex apps
Token limits & tooling reality
We talked candidly about token limits and how heavy builders hit them fast. Several strategies came up:
- OpenRouter for buying credits and routing across models
- Using different models for different tasks (Gemini for frontend, Claude for coding)
- Why open-source IDE agents move faster than locked tools
Skills, MCP, and the future of agent workflows
A big chunk of the conversation focused on Skills and why they matter.
- Skills load metadata first, full instructions only when triggered
- This enables progressive disclosure and smaller context windows
- Skills act like a standardized “dotfile protocol” for AI tools
- MCP handles tools; Skills handle instructions
Product ideas sparked
- Browser plugins that allow AI-driven changes within constraints (e.g. bulk calendar color rules)
- Proposal & marketing document tools where users say: “Move the logo, add padding, center it” — no massive UI needed
- Using Skills as customizable, client-specific branding & layout engines
A powerful habit: prompts as source code
One standout practice:
Treating prompts like source code history makes projects reproducible instead of “magic.”
Discussion question
Do you keep a prompt or change log for your projects?If not, how do you reproduce or debug AI-generated work?
Topics discussed (chronological)
- Publishing a first live IIS + SQL Server web app
- AI deployment guidance vs real-world system knowledge
- Progress through the Thumbio course
- Legacy background: COBOL, punch cards, System/360
- Transition into exec roles and back to hands-on building
- Hackathon plan: re-creating a VC-funded app via vibe coding
- Token limits and cooldown frustrations
- OpenRouter and model routing strategies
- Roo Code and open-source agent tooling
- Modes vs Skills
- Progressive disclosure and context control
- Skills as a standardized instruction format
- AI-driven customization inside constrained apps
- Document & proposal generation via chat-based edits
- Skill builders and marketplaces
- Prompt + change logs as reproducibility tools
- Why identical prompts can still produce different results
- Observability tools for tracking AI behavior