How people lose weeks of Claude Code work, and the four ways to not

plyo_dev — Thu, 03 Sep 2026 20:02:25 +0000

A pattern from the Claude Code and Cursor subreddits: someone builds for three weeks, asks for one more feature, and the AI rewrites half the project. The app breaks, the working version is gone, and the thread title is some variant of "how do I get my code back?" Usually they can't.

This post is the prevention list, from someone who reads those threads daily. I work on a product in this space (disclosed below), but most of what follows needs no product at all.

Why this keeps happening

AI coding tools edit files in place. Unless something is saving copies, there is exactly one version of your project: the current one. The AI has no undo across sessions, and "can you restore what we had yesterday" gets you a confident reconstruction from its memory of the conversation, which is not your code and rarely runs.

Option 1: learn just enough git

Git is the versioning tool programmers use, and you need three commands, not a course. In your project folder, ask your AI assistant to run them for you and explain each one: git init once, then git add -A && git commit -m "works" every time the app is in a state you'd be sad to lose. Restoring an old version is where git gets genuinely hostile to beginners, so also ask your AI to handle that moment when it comes. Cost: free. Risk: the restore step is exactly when panicked people make things worse.

Option 2: GitHub Desktop

A point-and-click app that does the above with buttons and shows you what changed between versions. No terminal. You still have to remember to commit, and restoring still involves concepts (branches, reverts) that bite under stress. But it's free, mature, and made by GitHub.

Option 3: copies of the folder

Zip the project folder when things work, name it with the date. Genuinely fine for small projects, scales terribly, and you will forget on the week it matters. If you do nothing else, do this.

Option 4: automatic checkpoints

This is the category I work in: Plyo connects to your AI over MCP, and every work session becomes a checkpoint on a timeline without you remembering anything. Restore is one click and never deletes history, so a bad restore can itself be undone. It also publishes your app online with safety checks. Free for three projects. The honest trade-off versus git: less control, less to learn, and you're trusting a product rather than a universal tool.

The one habit that beats every tool

Whichever option you pick, the failure mode is the same: saving only when you remember, which means never during the exciting parts. So anchor it to the AI instead of yourself. Literally add to your instructions file: "after any change that works, commit it" (git) or let checkpoints happen automatically (Plyo).

Where to put your AI-built app online when you don't know git

plyo_dev — Thu, 03 Sep 2026 20:00:54 +0000

Every week someone posts a version of the same question: "Cursor built my app, it works on my laptop, how do I get it online?" The answers they get usually assume knowledge they don't have. This is a map of the real options for people who built something with AI and have never used a terminal.

What "hosting" actually means

Your app currently runs on your computer, so only you can see it. Hosting means a copy runs on a machine that never sleeps, with a public address. That's the whole concept. Everything below is a different way of getting your files onto such a machine.

If your tool has a deploy button, start there

Lovable, Replit, v0 and Bolt can host what they build. One button, no export. The catch is lock-in of workflow, not code: your project lives inside that tool, and if you outgrow it (or its pricing) you'll be doing this exercise anyway. Perfectly fine for a first version.

If you built with Claude Code or Cursor, you have real files and more choices

Netlify Drop (drop a folder on a web page) is the fastest way to put up a static site, meaning a site with no database and no login. Free, no account needed for a trial. Its limit is exactly that: the moment your app has users, accounts, or saved data, dropping folders stops working as a way of life.

Vercel, Netlify, Railway, Render are what professional developers use, and they're genuinely good. The honest caveat for non-coders: they assume you can push to GitHub, set environment variables, and read a build log when a deploy fails. If those words mean nothing, you can still learn (plenty do), but budget an evening of frustration and keep your AI assistant open for the error messages.

Plyo is the one I work on, so read this knowing that. It's built for exactly the person this post addresses: your AI connects to it directly (over MCP, one paste), every version of your project is saved automatically, and publishing is one click, with a check for exposed API keys and broken pages before anything goes live. If a publish would break your live site, it doesn't go out. Free tier, three projects, no card. Where the professional platforms give you power and expect competence, Plyo gives you fewer knobs and expects none.

A VPS (renting a raw server) is the "full control" option lists love to include. Skip it for now. It's a part-time job, not a hosting choice, and the failure modes (security patches, full disks) land on you at 2am.

How to choose in one minute

Static site, no user data: Netlify Drop, done. Built inside Lovable or Replit and happy there: the built-in button. Real app with a database, built with Claude Code or Cursor, and you don't know git: Plyo. Comfortable with GitHub or wanting to learn professional tooling: Vercel or Railway.

Whatever you pick, do one thing before you share the link: search your own code for "sk-", "key" and "secret", or use any free scanner, because AI tools paste real credentials into frontend code more often than anyone admits. That mistake costs actual money and no host will catch it for you (checking for it on every publish is, disclosure again, one of the reasons Plyo exists).

DEV Community: plyo_dev