Show Notes
Parker dives into real-world Claude Code agent workflows, tests Whisper Flow’s action-oriented prompts, and sketches a practical path to turn daily notes into MDX-ready content with MCPs. It’s blunt, hands-on, and focused on what actually moves the needle.
Claude Code Agents: tailor, test, repeat
- Use case split: personal vs. project agents.
- Top setup: a “Full stack TypeScript engineer” agent handling Tailwind, Convex, shadcn, RX.
- Prompts: leverage Claude prompts CC and CCT to steer the agent; tailor prompts to your stack rather than grabbing templates from everywhere.
- Key takeaway: cook your own sauce. Don’t absorb every list or boilerplate—edit and adapt to your own projects.
- Practical tip: test agents in the exact context you’ll use them; start with a core stack and iterate.
Whisper Flow: action-oriented prompts and file tagging
- Whisper Flow brings action-based prompts into voice workflows, with faster, self-rolled inference.
- File tagging idea: reference files by full path or use tag/at tag commands to auto-tag or link assets.
- Real-world friction: some commands don’t work consistently across editors/environments (noted during testing).
- Takeaway: action-capable prompts are powerful, but expect rough edges and be ready to troubleshoot per workspace.
What not to assume about AI prompts
- Not every ambitious prompt (e.g., ASCII art generation) will work in Claude Code.
- Don’t rely on Claude Code for creative tasks that require specialized tooling; use it for what it does well and pair it with dedicated tools for art or visuals.
- Core lesson: prune experiments that don’t move the project forward; keep the focus on tasks the model handles well.
Content workflow: MDX, MCPs, and Notion
- Vision: use agents + MCPs to produce MDX-ready content (with components like code blocks and codelock).
- Intake: collect ideas and notes in Notion; MCPs format Markdown to MDX.
- Output: publishable blog content with inline MDX components; aim for a repeatable pipeline rather than one-off drafts.
- Practical example: a TypeScript-focused article might generate structured blocks, then render as MDX in your blog system.
Notable prompts and experiments
- Rubber duck prompt: a playful, punchy way to generate programming-moccomm prompts and copy for social. It’s workable but needs polishing.
- Twitter draft writer test: promising concept, results vary; you’ll need to tune tone and context.
- “Notion post effects”: building a pipeline to convert insights into MDX-ready blocks; keep it modular so you can swap components without breaking the whole flow.
Q&A highlights (condensed)
- HLD vs. LLD: High-Level Design and Low-Level Design—common software planning terms you’ll encounter.
- Amazon six-pager approach: a reference framework Parker mentions in context of design/docs.
- Background tasks: currently not the focus; stay practical and ship the core rocks first.
- Mason jar analogy (rocks, pebbles, sand): prioritize core, stable work (rocks) before layering on smaller enhancements (pebbles, sand) via background agents.
- Notion vs GitHub/blog publishing: focus on getting the “rocks” in place first (stable content and workflow) before overhauling the publishing stack.
Quick actionable takeaways
- Build stack-specific agents: start with one strong, stack-aligned agent, then iterate.
- Don’t clone every prompt from lists; tailor them to your pipeline and output requirements.
- Use Notion as a lightweight intake for ideas; use MCPs to format into MDX for your blog.
- Keep a simple priority: rocks first (core content and workflows), pebbles and sand later (UX polish, additional features).
Links
If you’re using this daily series as a quick starter, focus on one agent tweak, one MDX workflow tweak, and one Notion-to-MDX improvement this week.