Show Notes
Today Parker talks through building your own AI-enabled template, why a DIY stack beats chasing a ready-made one, and how a Python + Next-inspired approach could mirror the T3 vibe without sacrificing practicality.
Build-your-own-template philosophy
- Don’t buy a template. Build your own so you learn faster and it fits your needs.
- Inspiration is fine, but you gain the real benefits by doing the work in your own codebase.
AI SDLC: the updated prompt flow
- Updated AI SDLC flow adds clarity and context. It’s a sequential, manual prompt chain that you treat like a real product team would.
- Key prompts and flow elements (8 of 14 covered):
- Start with idea and solution; require three solutions
- PRD + architecture
- System patterns, tasks, and a new Task Plus prompt
- GitHub-style instructions, dependency clarity
- Atomic implementation details and self-contained documentation
- Context is king. Augment-like context engines help by injecting codebase context into prompts.
Business prompts and opportunities
- OpenAI paper reference: lessons from frontier companies and how to surface concrete opportunities using AI workflows, dev tooling, product management, teaching, and community building.
- Ideas include a plug-and-play evaluation harness and “AI inside product studio” to embed AI into products.
Augster, JSON prompts, and prompt storage
- Augster: a promising prompt project with ongoing updates (XML-based prompts are being examined).
- JSON prompts and the challenge of saving prompts reliably; multiple formats and tools exist.
- Caution on overkill: Prompt Methus (prompt engineering IDE) can be too heavy; aim for practical, lightweight storage and reuse.
Auggie GA and remote agents
- Auggie GA is coming; Parker plans a workflow to map job descriptions and teams to agents.
- The goal: orchestrate agent-driven inquiry and automation for real projects.
Architecture and stack: pragmatic, composable design
- Architecture sketch: CI/CD pipeline, Debian VPS, Docker Compose, container registry (GCR), Nginx, and a server/API layer.
- Front-end: Next.js app (app router) with selective FastAPI usage for non-user-facing services (e.g., payments, Google integrations, AI embeddings).
- Discord bots are integrated via the back end; observability tooling (Medi) feeds back into Discord.
- Core idea: build a composable stack that agents already know well—Next.js, FastAPI, and Supabase.
The T3-inspired path: F3N stack
- The concept: a Python-centric take on the T3 stack with strong type safety across Python and TypeScript.
- Proposed stack skeleton (F3N):
- FastAPI + OpenAPI-generated types
- Python side for models and business logic (Pydantic)
- TypeScript + OpenAPI-generated types on the TS side
- Supabase + SQLAlchemy (instead of Prisma/Drizzle)
- API layer type safety: generate and share types between Python and TS; reduce drift.
- Practical flow:
- Define models in Python
- Generate TS types from OpenAPI schemas
- Build a type-safe API client (TRPC-like feel) for the frontend
- Example focus: a mutation to update a user profile biography, with field-level control and type-safe updates.
Practical next steps to build your own template
- Start with a stack you like (T3-like patterns) and map them to Python + TS.
- Use a design where the API surface is type-safe across both languages.
- Build a small PoC (e.g., a profile update flow) to test the typing and client wiring.
- Keep the template composable: swap in agents, dashboards, or AI services as needed.
Quick takeaways
- Learn by building your own template; avoid locking yourself to someone else’s workflow.
- Use a T3-inspired, Python-friendly stack to get type-safe cross-language APIs.
- Focus on context, composability, and practical prompts that actually drive code and decisions.
Links
- Create T3 App - The T3 Stack for full-stack typesafe Next.js apps
- Supabase - Open-source Postgres database platform
- Pydantic - Python data validation library
- FastAPI - Modern Python web framework with OpenAPI support
- Augment Code - AI coding assistant with codebase context
If you found value in mapping ideas to a concrete, composable template, drop a like, join the Discord, and subscribe for the next update.