Show Notes
In just two days, Parker rebuilt a SaaS using AI and a practical, just-in-time learning framework. Here are the core lessons, the stack, and the playbook you can steal.
Key takeaway
- Learn by doing, with a deliberate, just-in-time process: write down annoyances, generate ideas, then prototype quickly to uncover what you actually need to learn.
The learning framework in practice
- Capture irritations and problems that bug you daily.
- Build a big backlog from those notes and prune into learning goals.
- Use curated learning lists (AI engineering topics, deployment patterns, etc.) to pick where to start.
- Create a reference codebase in a temp folder to compare against your current work.
- Use Augment (or similar tooling) to run comps, expose gaps, and generate a list of concepts you don’t yet understand.
- Start with a tiny, end-to-end prototype and scale toward a full MVP.
- Stay humble with LLMs: ask questions, don’t try to impress the model—learn from it.
Architecture snapshot (2-day rebuild)
- Frontend: Next.js (LM experience) replacing Astro for remote-agent workflows
- Backend: FastAPI
- Database/Auth: Supabase with PostgreSQL
- Payments: Stripe (with webhooks)
- Messaging/Orchestration: Discord bot serving as a front-end for agent stuff
- AI/Indexing: Gemini Pro, Vertex as a search/indexing layer
- Data/source management: Bright Data (data marketplace) for scraping without API keys
- Infra/Deployment: Netlify (serverless), Nginx reverse proxy on Debian, containers
- Observability/ops: OpenTelemetry-inspired ideas; “self-healing” agentic workflows concept (Dino as a reference point)
- Other integrations: Google ADK for learning paths, notional integration with Mermaid, Notion, Linear, GitHub
Learn-by-building decisions that drove progress
- Replaced older front-end stack with Next.js to better support LM-driven workflows and remote agents.
- Built a FastAPI backend that orchestrates AI ops and talks to external services (Stripe, Discord, DB, etc.).
- Implemented a Discord bot as the testbed front-end for agent capabilities and project workflows.
- Used a “membership/site scaffold” approach for VI, including a learning-path backbone, to test end-to-end onboarding and payments.
Product and learning takeaways
- Product management is shifting: test features quickly with a Discord bot and a lightweight registry endpoint before full-blown UX.
- A well-organized learning stash accelerates progress: use curated lists, “vision agent” concepts, and OpenAI/Google agent SDKs to jumpstart projects.
- Versionable, reusable learning code: keep a developer folder with temporary codebases to compare ideas and expose gaps.
Practical tips for viewers
- Keep a dedicated development folder with a tmp subfolder for reference codebases you’re learning from.
- Drop your reference project into your active workspace and use augment-like tooling to surface missing concepts.
- Build in public or semi-public, but focus on the learning and the process—not just the final product.
- Use Discord as a lightweight front-end for agent-oriented experiments to accelerate iteration.
Learning resources & prompts (examples mentioned)
- AI engineering and agent workflows lists
- Vision agents and Python AI SDKs
- OpenAI agents Python SDK (Google’s AI tooling)
- Awesome lists (general resource hub for prompts, frameworks, libraries)
- Prompts and prompt-learning collections (Discord-centric prompts, etc.)
- Meridian project (idea for delivering personalized, concise briefs)
- Morphic, FFUF, Perplexity, Auditor (learning projects/examples)
Actionable next steps for you
- Start your own learning backlog: write down daily annoyances you want solved with automation.
- Create a small reference codebase in a temp folder and experiment with augment-like tooling to identify gaps.
- Pick one end-to-end idea (e.g., a Discord bot that orchestrates a simple agent task) and prototype it in a weekend.
- Organize a small learning shelf: 2–3 curated lists, one learning path, and one quick prototype guideline.
Links
- Augment (learning/automation tool)
- Gemini Pro
- Next.js
- Netlify (serverless)
- Supabase
- Stripe
- Discord
- Vertex AI (search/indexing)
- PostgreSQL
- Nginx
- Debian
- Bright Data
- OpenAI API / Python SDK
- Google AI SDK / learning paths
- Mermaid, Notion, Linear, GitHub (integrations and tooling)
- Morphic, Perplexity (learn-by-reading/examples)
- Awesome lists (prompts, libraries, frameworks)
If you found a takeaway you can apply today, drop a comment and try one actionable item this week.