AI Frontier Fridays — Session 1: G-Stack
AI Frontier Fridays — Session 1: G-Stack
Date: Fri, 22 May 2026 Presenter: Anuar Ustayev
Attendees: Anuar Ustayev, Monika Popova, Aleksandra Rubaj, Osahon Okungbowa, Daniela Popova, Almubdi Mutaikhan, Luccas Mateus, Yoana Popova, João Demenech, Rufus Pollock
Links
Summary
First session of AI Frontier Fridays — weekly experiment for the team to share AI setups and workflows. Live demo of G-Stack by Gary Tan (YC CEO): an open source AI agent team management system (100k+ GitHub stars) with 20+ tools serving roles like CEO, design engineer, QA, release manager. Designed for Claude Code but works with other LLMs.
Installation & Setup
- Took ~5 min (advertised 30 sec)
- Requirements: Claude Code, Git, Bun (auto-installed), Node.js
- Team mode creates a shared
.claude.mdfor repo-level collaboration — others can clone and use skills without installing individually - Office hours skill required a Claude session restart to activate
- Setup includes telemetry and cross-project learning options
What we built
Used G-Stack Builder mode to design a project live:
- Selected "having fun" mode and content series as project type
- Defined target: non-technical people curious about AI, Friday cadence, demo-first format
- G-Stack conducted landscape research and generated premises — confirmed all 5
- Evaluated 3 architecture options; team voted for Approach C: notebook-first (Observable/Jupyter-inspired)
- Design doc created → adversarial spec review run → 8/10 quality score → approved
- Selected interactive data visualizations as the first demo topic
- Scaffolded a Cloudflare Worker + Anthropic API integration
- Successfully generated a bar chart: global coffee consumption by country
- Site partially functional — missing CSS styling; paused for now
Decisions
- Mode: Having fun (not startup/hackathon)
- Project type: Content series — AI Frontier Fridays
- Vision: Curated archive of remixable experiments; each Friday post as a reusable building block
- Architecture: Approach C — notebook-first (Observable framework)
- Design doc: Approved at 8/10
- First demo topic: Interactive data visualizations using AI
G-Stack vs Superpowers AI
João shared a benchmarking comparison from X:
| G-Stack | Superpowers AI | |
|---|---|---|
| Orientation | Product builders | Engineers |
| Approach | Structured, guided, rigorous | Faster, less guided |
| Output | Multi-page app (Next.js) | Single-page app |
| Criticism | Production readiness concerns (loading test files) | Tests often "happy path" and useless; can be intrusive |
Aleksandra noted G-Stack steered her side project toward engagement features rather than her simple custom requirements. Anuar disabled Superpowers when using Gas Town (conflicts with mayor's own planning + beads workflow).
Next steps (from Gemini notes)
- Anuar — Pick first demo topic: research interactive data viz concepts suitable for non-technical audience
- Anuar — Add Anthropic API key to G-Stack config file
- Anuar — Push G-Stack repo to GitHub for team collaboration
- Anuar — Share recording with absent team members
