← Digesting

AI in my workflow

Filed in: AI

These are my /ai notes on how I use (or don't use) AI here. The format is inspired by the slash pages idea.



I use AI in my writing workflow, far more than I'd have predicted a year ago. Here I'm trying to be honest about what's actually working and what isn't.

My approach keeps evolving. Posts remain primarily me-authored; AI helps with capturing ideas, roughing out structure, and pressure-testing arguments.

A common writing workflow now#

TL;DR: I capture ideas quickly, structure with AI, write the content myself, then use AI to pressure-test and format.

I do this partly to learn about the tools and their limits. AI-generated outlines usually need restructuring. Generated content always needs checking.

Tools I'm using and interested in#

Tool Type Status Description
Claude Code AI coding agent Current CLI agent for longer coding sessions on this site. Project memory, custom skills, and hooks keep it aligned to site conventions without repeating context each time.
GitHub Copilot AI coding agent Testing Exploring the terminal interface for coding tasks.
Cursor IDE Past AI pair‑programming IDE built on VS Code. Replaced by Claude Code for most tasks.
Fireflies Transcription Current Meeting recorder and transcription with integrations for Zoom, Teams, and Google Meet. Quick search, summaries, and action items across calls.
Highlight.app Desktop assistant and audio transcription Current Lightweight desktop recorder for capturing quick audio notes from calls. Fast, friction‑free capture; mostly for audio memos, pushing toward structured snippets.
Manus AI agents Testing AI agent platform for multi-step tasks. I find this is exceptional at drafting HTML slides that can then be imported to PPT for refinement.
Perplexity AI agents Current Free annual trial; good multi‑tool search and experimentation.
VEED Video/captions Future Audio‑to‑captions and filler‑word trimming (ums/ahs) are excellent for quick polish on short clips. I don't have a regular need right now, but it's on my shortlist when I do.
Claude Code agent profiles Code agents Future Experimenting with custom sub‑agents (refactorer, security auditor) to tighten prompts and roles. Using community profiles from Claude Code Custom Agents repo.
Microsoft Copilot Studio Assistant builder Testing Certainly not my first choice, but in a Microsoft ecosystem it unlocks useful integrations: grounding assistants on SharePoint/OneDrive content, and experimenting with a parent–child chat agent dispatcher pattern for routing tasks.
Context7 MCP MCP Current MCP server for fetching library docs at query time — handy when working with APIs I don't know by heart.
Browser MCP MCP Current MCP browser client for integrating local tools.
Chrome DevTools MCP MCP Current MCP server for Chrome DevTools automation and inspection.
Awesome ChatGPT Prompts AI agent prompts Current Open collection of LLM prompts.
Commet Browser AI assisted browser Current AI‑assisted browsing for research and summaries.
Together AI + FLUX.2-dev AI image generation Current Hero image generation via the FLUX.2-dev model, run from a browser tool built into this site. Prompt, generate three in parallel, crop, save. About $0.03 a post. See the post about it.
Loras AI image maker Past Fun and lightweight. Moved to FLUX.2-dev for more control over style and tighter workflow integration.

What's next#

The image generation workflow is now settled enough that I wrote a post about it. Next I want to see if the audio→draft pipeline can get tighter — there's still too much manual cleanup between transcription and a usable outline.

If you have a tool or approach that's working for you, I'd love to hear it.


More things I've been digesting