Writing
Lessons learned building AI agents, automating workflows, and navigating the AI landscape as a TPM.
Mar 14, 2026
Most AI-generated outputs vanish into terminal history. I turned Claude Code's memory folder into an Obsidian vault — now every draft, image, and debugging session is browsable, searchable, and backed up.
Mar 12, 2026
How I overhauled my Next.js portfolio in one session using a Framer template as design inspiration and Claude Code as the implementation partner — scroll-stacking cards, film grain, transparent navbar, and more.
Mar 9, 2026
What if someone could see your portfolio by typing ssh into their terminal? I built a 6-page terminal UI with Go, Bubble Tea, and Wish — featuring typing animations, ASCII art, and a markdown blog reader — and deployed it to Fly.io for $2/month.
How I turned 28 existing blog posts into a 4x/week LinkedIn content strategy — using a Python script, Late API's free tier, and a content transformation pattern that takes 0 manual effort per week.
How I built a multi-platform social media automation pipeline for AgnusBlast using Late API, Claude Cowork, and Imagen 4 — scheduling 32 posts across Instagram and X/Twitter with one script.
Mar 7, 2026
26 blog posts, zero images, one afternoon. Here's how I used Imagen 4 and Gemini 3 Pro to generate custom hero visuals for every post — with 8 different visual styles mapped to content type.
Mar 6, 2026
Who said you need a desk to ship? How I use Claude Code on claude.ai to brainstorm, write, and update this site from my phone — and where it still falls short.
Mar 5, 2026
A step-by-step guide to cloning, configuring, and deploying the ai-journey site — including CLI tools, database credentials, sanitization rules, and all the gotchas I hit along the way.
How I use structured prompts and Napkin AI to turn weekly progress data into visual infographics — turning reflection into a habit.
Mar 3, 2026
I was paying $24/month for n8n Cloud without realising I could run the exact same thing locally for free using Docker. Here's how Claude Code walked me through the whole migration — including auto-exporting all 24 workflows — in a single session.
Mar 2, 2026
I wanted to know the moment resale ticket prices dropped below my threshold — without checking manually. A Python script, Playwright, and a GitHub Actions cron job later, I never have to look again.
Mar 1, 2026
I built a trip planner with zero npm dependencies — plain HTML, CSS, and vanilla JS. It deploys in seconds, loads instantly, and taught me how much the web platform already gives you for free.
Feb 28, 2026
OCR alone gives you gibberish. Add Claude Haiku and suddenly 'DRGN FRT 2PK' becomes 'Dragon Fruit' with a 5-day shelf life and a purchase date pulled from the receipt header.
Feb 27, 2026
Feb 26, 2026
Feb 25, 2026
How the desk CLI eliminated auth failures and reduced token usage by 52% in our AI agent workflows.
Feb 24, 2026
I run Claude Code on my phone (cloud) and my Mac (local terminal). Here's what I learned about keeping them in sync — and when the quick fix on mobile costs more than waiting.
Feb 22, 2026
A practical guide to Model Context Protocol — connecting AI models to Jira, Confluence, Google Workspace, and enterprise tools.
Feb 20, 2026
From plumbing to intelligence — portfolio analysis, wiki hygiene, organizational knowledge automation, and the real lessons about when to stop automating.
Feb 18, 2026
Lessons from writing system prompts for 17 agents that run daily — specificity, failure modes, and the art of agent instructions.
Feb 16, 2026
How I built 17 specialized AI agents that autonomously handle sprint boards, meeting prep, and cross-team program monitoring.
Feb 14, 2026
After testing 9 AI tools side by side — from NotebookLM to Claude Code to Gemini — here's what actually works for program management.
Feb 12, 2026
The evolution from simple automation scripts to a sophisticated multi-agent system with Claude Code.
Feb 10, 2026
Long before Claude Code, before MCP, before agents — there was AutoHotkey. How scripting clinic workflows in 2012 planted the seed for everything that followed.
Feb 8, 2026
Replacing unreliable MCP data sources with direct CLI tools, implementing model tiering (Opus vs Sonnet), and cutting weekly costs from $333 to ~$200.
Feb 7, 2026
Building a custom statusline for cost visibility, hitting macOS quirks, attempting (and failing) to schedule agents, and learning to self-audit your config.
Feb 6, 2026
Adding program monitors for every workstream, building initiative health auditing, and discovering that Jira's custom fields are the key to everything.
Feb 5, 2026
Search indexing lag, silent failures, result pollution — Day 5 was about discovering that production-grade agents need production-grade error handling.
Feb 4, 2026
One agent to rule them all. How a hierarchical dispatch system turned compound requests into parallel agent execution.
Feb 3, 2026
Not all AI models are created equal. Day 3 was about learning which tasks need the expensive model and which don't.
Feb 2, 2026
Morning briefing, EOD summary, standup sync — three agents that turned 3 hours of daily TPM overhead into 5 minutes of review.
Feb 1, 2026
Installing Claude Code, writing my first CLAUDE.md, connecting MCP servers, and going on a skill-install spree that laid the foundation for everything that followed.