23d • General discussion
AI Developer Accelerator — Coaching Call - March 03
Meeting Purpose
Key Takeaways
- Claude as an Operator: Patrick's "Infrastructure as Context" workflow uses Claude to automate VM provisioning and management via tmux and the /remote-control command, enabling mobile control of multiple, isolated instances.
- Strategic Shift to Infrastructure: With software development commoditized, Ty is shifting focus to foundational research (e.g., non-transformer models, AGI testing) to find the next breakthrough.
- Client Communication is Key: To overcome non-technical client fears (e.g., Azure privacy), use simple analogies and frame solutions as the client's own idea to build trust and ownership.
- Obsidian + Claude is a Powerful Stack: This combo creates a private, local "second brain." Claude in an Obsidian terminal provides direct, context-aware access to all notes, avoiding the cost and complexity of a vector database.
Topics
Claude as an Operator: "Infrastructure as Context"
- Patrick's workflow automates VM provisioning and management using Claude as the operator. Process: A single architecture document (the "context") is fed to Claude, which then configures a blank Ubuntu VM. Automation: This includes hardening, SSL setup, Docker deployment, and installing Claude Code itself. Remote Access: The /remote-control command is run inside a tmux session, creating a persistent, mobile-accessible CLI. Benefit: This method bypasses Claude's one-session-per-machine limit by creating multiple isolated VMs/containers, each with its own operator.
- Real-World Application: Patrick used this setup to refactor his intel.patchutech.com project in 30 minutes for a client, demonstrating the efficiency of "Infrastructure as Context" over traditional scripts.
Strategic Shifts & New Projects
- Ty's Focus: Shifting from application development to foundational research (e.g., ARC AGI-1/2 testing, non-transformer models). Rationale: Software development is becoming commoditized; the next breakthrough will be at a deeper, architectural level.
- Marc's Goal: Gaining hands-on experience with model training. Project 1: A resume ranker (forward/hold/reject) using Google Colab and Hugging Face. Project 2: A movie recommender trained on user preferences. Recommendation: Use small models (e.g., Gemma 1B) for learning. They are cheaper and faster, and their failures more clearly expose training weaknesses.
- Ryan's Projects: Video Platform: Fixing streaming issues for clients with poor internet by implementing adaptive resolution (Cloudflare Stream). Alternative: Use ffmpeg scripts generated by Claude for more control. Lead-Gen SaaS: A new lead-gen app using Meta Glasses to capture business van info, which an AI then researches to generate outreach emails.
Client Communication & Privacy
- Problem: Andrew's expert witness app faces client resistance over data privacy in Azure. Issue: Clients trust OneDrive (M365) but are wary of other Azure services, even within the same tenant. Root Cause: A perception issue, not a technical one, driven by a lack of IT expertise.
- Solution: Use simple analogies to build trust. Example: An Azure tenant is a secure "walled-in office." OneDrive is the "desk," and other services are the "closet"—all are within the same secure space. Tactic: Frame the solution as the client's own idea to build ownership and reduce resistance.
Training & "Second Brain" Infrastructure
- AI Education: To teach AI, focus on the skill of detailed, conversational prompting. Rationale: AI requires context, which is best practiced by explaining a project to a human collaborator. Method: Use AI for tasks that automate non-core skills (e.g., CV writing), allowing users to learn why changes are made.
- "Second Brain" Infrastructure: Recommendation: Use Obsidian with a Claude Code terminal plugin. Rationale: This creates a private, local "second brain" with direct, context-aware access to all notes, avoiding the cost and complexity of a vector database. Syncing: Use a GitHub repo for free, automatic synchronization across devices.
Next Steps
- Elena: Review last week's meeting recording for fundraising resources.
- Marc: Continue model training projects; consider using a smaller model (e.g., Gemma 1B) for learning.
- Ryan: Implement adaptive video streaming; research ffmpeg as an alternative to Cloudflare Stream.
- Andrew: Use analogies to explain Azure privacy to clients; evaluate Azure Content Understanding.
- Elijah: Experiment with the Obsidian + Claude Code setup for a "second brain."
Action Items
- Install Claude mobile app; try /remote-control - WATCH (5 secs)
- Review Brandon's fundraising recording; update group - WATCH (5 secs)
- Start Cloudflare Workers project (Astro+Wrangler) - WATCH (5 secs)
- Share Fieldy pin link in chat - WATCH (5 secs)
- Evaluate FFmpeg for video transcoding; integrate if viable - WATCH (5 secs)
- Propose Shipkit training to employer; explore funding - WATCH (5 secs)
- Add case studies/projects to landing page - WATCH (5 secs)
- Share example no-scroll web app link in chat - WATCH (5 secs)
- Implement Discord buttons to forward alerts to Claude Code on server - WATCH (5 secs)
- Draft 10-yr-old analogies for Azure tenant/privacy; use in client comms - WATCH (5 secs)
1
3 comments
AI Developer Accelerator — Coaching Call - March 03
skool.com/ai-developer-accelerator
Master AI & software development to build apps and unlock new income streams. Transform ideas into profits. 💡➕🤖➕👨💻🟰💰
Leaderboard (30-day)
1
+28
2
+10
3
+10
4
+5
5
+4
Powered by