Uses

My primary code editor. The tab completion and Composer mode changed how I build. I use it for almost all code I write.

For longer, autonomous coding tasks — architecture decisions, refactors, and overnight work delegated to my AI assistant Matua.

My default framework. React Server Components, ISR, and App Router make it the right call for almost every project right now.

Postgres + Auth + Storage in one. Real-time subscriptions are useful for workout tracking. RLS is non-negotiable for multi-user apps.

Every project deploys here. Preview deploys per branch are the feature I couldn't live without.

Fast, consistent UI without fighting CSS. shadcn/ui gives me composable, accessible primitives I actually own.

For animations that make apps feel alive without overcomplicating things. layout animations are underrated.

My primary AI model. Better reasoning and writing quality than anything else I've used. Also powers Matua, my AI assistant.

My AI assistant runs on a Hetzner VPS via OpenClaw. Overnight work, cron jobs, proactive monitoring — it wakes up while I sleep.

For Codex — the OpenAI coding agent — which handles code changes while I handle architecture and review.

Quick research and fact-checking. Better than Google for developer questions because it gives synthesised answers.

Stages all changes, sends the diff to Claude, gets 3 conventional commit options, arrow-key pick. Saves real time every session. Makes commit history actually readable.

Branch diff vs main → structured PR description with summary, change breakdown, and testing notes. Used before every pull request.

Pipe any code file to Claude → plain English explanation. Modes: normal (full walkthrough), --brief (one paragraph), --debug (explanation + issues + suggestions).

Pre-push code review grouped by severity 🔴🟠🟡🟢 with a code quality score 0-100. The habit that catches issues before they become PRs.

git log → Claude groups commits into a formatted changelog (Features / Fixes / Refactors). Used for GitHub releases and portfolio blog posts.

Generates a complete AI-ready context document for any repo: recent commits, git status, stack, routes, source files, README, env vars, TODOs. Perfect for handing off to sub-agents.

Scans all git repos, reads commits from the last N hours, sends grouped data to Claude Haiku, returns a professional standup in 4 seconds. My actual morning workflow.

Powers the Ship Log and Activity pages on this site. Free, no auth needed for public repos, plenty of rate limit for a personal portfolio.

Free, open-source weather API with no API key required. Used in NZ Adventure Planner and Snow Forecast apps.

For interactive maps. The GeoJSON layer API is more performant than DOM markers at scale.

FIT file parsing for the Durability Analyzer. Client-side only — no data leaves the browser.

Daily driver. M-series chip makes local dev fast enough that I rarely wait for builds.

4 vCPU, 8GB RAM, US West. Runs my AI assistant Matua full-time. Cheaper than leaving a Mac on 24/7 and more reliable.

Connects Mac to VPS over a private network. SSH locked to Tailscale only — no exposed public SSH port.

Task board and notes. My AI assistant checks it for high-priority tasks and executes on them overnight.

How I communicate with Matua. Morning briefings, task delegation, overnight summaries — all via Telegram.

Everything is committed and pushed. I use two accounts: harry-supermix for Harrison's projects, matua-agent for independently built apps.

Wireframing before building. Even rough lo-fi sketches save time when the AI agent is generating code.

Colour palette exploration. I tend toward muted, high-contrast palettes that work in both light and dark mode.