Uses

What I actually use

Short and opinionated — the reasoning matters more than the list.

Editor & terminal

VS CodeExtension ecosystem for Python and web work outweighs a faster-but-narrower editor for daily client work.

Windows Terminal + WSL2 / Git BashLinux tooling for the backend, native Windows for everything else — fewer surprises than dual-booting.

zsh with a minimal promptFast startup matters more on a machine that's also running a WebGL dev server and a model API in parallel.

Machine

Mid-range laptop, dual monitor at the deskDeliberately not top-spec — if a 3D scene or a demo is smooth here, it's smooth on what a hiring manager is using too.

AI tooling

Claude CodeAgentic terminal workflow fits multi-file refactors better than a chat window with copy-paste.

GitHub CopilotInline completion for boilerplate; not trusted for architectural decisions.

Model providers benchmarked

Anthropic ClaudeBest instruction-following for multi-step agent prompts in testing so far.

OpenAI GPTStrong fallback and still the widest tool-calling compatibility across client stacks.

Open-weight models via self-hosted inferenceOnly where a client requires data residency the hosted APIs can't satisfy — the trade-off is ops overhead for control.

Hosting

Fly.ioGlobal edge deploy with WebSocket/WebRTC support that a plain PaaS often complicates.

Pipecat CloudPurpose-built for real-time voice pipelines — reinventing that infrastructure per client isn't billable value.

GitHub PagesZero-cost static hosting for a site with no backend of its own; the trade-off is no server-side headers, hence the _headers file documented for migration.