Claude Code is the one I keep in the terminal. It is useful on a browser app like ours because most of the work is plumbing rather than clever algorithms: wiring a WebRTC offer/answer handshake, adjusting a state hook, chasing a race in a connection handler. I let it read the relevant files first and ask for a small diff at a time, then run the app myself before I trust it. The times it goes wrong are when it guesses at library behaviour instead of reading the installed version's docs, so I paste the docs in.
I use ChatGPT for this rather than a dedicated paraphraser, because I can paste the paragraph and say what tone I want in the same message. What works for me is asking for the rewrite one paragraph at a time and keeping the original open next to it, so I catch the places where the meaning drifts. The failure mode I watch for is it swapping a precise term for a vaguer synonym, which reads smoother but says less.
Sentry is what I reach for on a small web app. The part that earns its keep is the stack trace with the browser and OS attached, because half of our bugs only ever happen on one specific phone. I keep the free tier and set the sample rate low so the quota lasts; the alert email plus the release tag is enough to tell whether a deploy introduced the error. It is heavier than I need for a static site, but for anything with a login it pays for itself.
For moving a few big video files between my own machines I use rsync over SSH when both ends are on my network, and AirDrop when it's phone to Mac. When the other person isn't technical I fall back to WeTransfer's free tier, but the 2 GB cap and the upload wait are the usual annoyance.