Show Notes
Parker walks through planning a cloud-code workflow using Claude Code and parallel agents to automatically assemble a daily whiteboard from X (Twitter) lists and bookmarks, with a practical implementation plan and tradeoffs.
Claude Code planning approach
- Uses Claude Code with Ultra Think and parallel agents to analyze a problem, surface options, and generate a structured plan.
- Think in steps: trigger extended thinking with keywords, spawn sub-agents to evaluate parts of the problem, and converge on a succinct plan.
- The process yields a concrete task list and architectural choices, not just ideas.
Daily whiteboard system: goals and flow
- Build a system that auto-populates a whiteboard for each day from:
- Public X bookmarks lists
- Public X posts added to specific lists
- Data path:
- Detect new bookmarks or list updates
- Store links in a database (SQLite preferred for small scale)
- Render on a per-day whiteboard in a Next.js app at
/whiteboard/{day}
- Architectural note: prefer a public access path to bookmarks/lists, with a headless browser if needed to bypass API limits.
Architecture decisions and tradeoffs
- API vs scraping:
- X API has limited free reads; scraping is preferred to avoid monthly API costs.
- Database choice:
- SQLite with Drizzle ORM for quick, lightweight storage; easy to migrate if traffic grows.
- Presentation layer:
- Excaliraw whiteboard (open source) with a possible fork for per-day boards.
- Hosting and execution model:
- Next.js on Vercel (serverless constraints) vs a dedicated server.
- If real-time listening is needed, a listening service (not just cron) is favored over infrequent polling.
- Scraping tooling:
- Playwright as the scraping tool to log in and fetch bookmarks/lists when API access isn’t viable.
Implementation plan (step-by-step)
- Project setup
- Create a new daily upload whiteboard directory in the dev repo.
- Establish a Next.js route structure for daily whiteboards: /whiteboard/[day]/page.tsx.
- Data sources and access
- Identify the public X bookmarks list path: daily/upload q.
- Decide between watching lists vs individual bookmarks; plan for both triggers.
- Database layer
- Set up SQLite database with a simple schema for links, sources (list or bookmark), and day/date.
- Integrate Drizzle ORM for query convenience.
- Scraping/ingest engine
- Implement a scraper (Playwright) to fetch links from public bookmarks and lists if API access isn’t feasible.
- Build a pipeline to push new items into SQLite and trigger a whiteboard render.
- Whiteboard rendering
- Integrate Excaliraw to render daily boards; fork if customization is needed.
- Auto-generate a per-day board at
/whiteboard/{day}.
- API routes and workflow
- Create Next.js API routes to surface content to Excaliraw and manage the pipeline (content-to-board).
- Event triggers and freshness
- Prefer a listening approach over fixed cron intervals for real-time updates (avoid 6-hour latency).
- Feasibility and pricing checks
- Re-check X API pricing and rate limits; if scraping remains necessary, prepare guardrails to stay compliant.
- Final plan and options
- Document the preferred path (X scraping + SQLite + Excaliraw) with tasks, owners, and estimates.
- Review and iterate
- Use Claude-based planning to re-scope, add options, and prune low-value tasks.
Key takeaways
- Leverage Ultra Think and parallel agents to rapidly map a complex integration plan.
- Start small: SQLite + a single day’s whiteboard to validate the pipeline before scaling.
- If API limits bite you, design for scraping with clear compliance and robustness considerations.
- A per-day whiteboard keeps content organized and makes the automation tangible.
Q&A and community notes
- Questions touched on intro-to-AI coding foundations, documenting procedures, and building a content calendar.
- Suggestions include outlining a procedural overview video and sharing a transparent workflow tracker.
Next steps and callouts
- If you want to join the weekly VI call, join the Discord community (engineers from Microsoft, Google, etc. participate).
- Follow along for updates on cloud-code strategies and deeper dives into parallel agents and planning workflows.
Links
- Excalidraw (whiteboard rendering): open-source whiteboard rendering for daily boards
- Claude Code: planning and reasoning with parallel agents
- Playwright: browser automation for scraping
- SQLite + Drizzle ORM: lightweight database stack for small projects
- Next.js: framework for the daily whiteboard app
- X / Twitter API pricing and limits: developer portal and rate limits
- Bear notes (CMS integration discussion)