Blog
Deep dives on AI-generated codebases, localization architecture, and how to keep your product multilingual without rewrites.
Lovable reads three kinds of instruction: workspace and project knowledge, on-demand skills, and instruction files in the repo such as AGENTS.md. Each is read at a different time, which is why a translated Lovable app slides back into English after a few prompts.
Read article →
Every localization ROI model assumes two numbers you do not have: a cost line big enough to amortise and a forecast somebody will stand behind. Without them the question is not what the return will be, it is what test you can run cheaply enough to find out. Three published cases, two of them from people working alone, say what that test should look like and where it goes wrong.
Read article →
Lovable approved globalize.now as a Select Partner on 6 October 2026. What the status means, how a Lovable app goes from one language to 50+ with one prompt, and what happened when RajaRide tried it both ways.
Read article →
Ten short i18n rules to paste into AGENTS.md, .cursor/rules, CLAUDE.md or a Copilot instruction file, with the reason behind each one and where every tool reads it. Then the limit: two of the code surfaces developers use most never read the file, and what enforces the rule when the file cannot.
Read article →
UI copy stays consistent when every user-visible string lives in one catalog in the repo, is referenced by key, and changes only through a reviewed diff. The practices that make that stick: catalog layout, key naming, copy review in pull requests, a lint rule for new literals, and a style-guide file your AI coding agent reads.
Read article →
Most growth advice assumes a team, and it never says so out loud. The assumption is buried in the last paragraph, where the tactic quietly turns into a workflow somebody has to own. Here is the test that sorts the tactics one person can finish from the ones that need staffing, and what the evidence looks like on both sides.
Read article →
On 29 September we shipped globalize.now in Japanese and Brazilian Portuguese: every page, every pricing table, all 61 blog posts. Translation took eight minutes, automated QA took 27 more, and the app billed about €31. Here is the receipt, what the QA flagged, what the engineering involved, and what we have not checked yet.
Read article →
Localizing a product to get more users is a decade-old tactic with one well-documented case behind it, and almost nobody finishes it. Here is what that case reported, what running it took at the time, who is still running it in this category, and what changed about the work.
Read article →
Base44 writes your interface text straight into React components, so the app ships in one language. Turn on GitHub sync, keep the translation catalogs in that repository, and let Base44 pull every merge to main back into the editor.
Read article →
globalize.now is live in Claude's connectors directory. Connect it once and run localization in plain language: translate locale files you drop into the chat, connect a GitHub repo, add languages, and set the glossary and style rules every translation follows.
Read article →
Bolt.new hands you a real Vite and React codebase with every string written into JSX. Push it to GitHub, keep the translation catalogs as committed files, and nothing third-party sits in your render path.
Read article →
Lovable workspace skills are markdown playbooks the agent loads when a task matches. Import the lovable-i18n skill once, add the globalize MCP, and Lingui catalogs land in your own repository instead of a dashboard.
Read article →
Replit Agent writes UI text straight into components, and asking it to translate the app gets you a half-result that drifts. The repository already exists as checkpoints: connect it, keep catalogs in Git, and pull the translations back into Replit.
Read article →
Next.js 16 renamed middleware.ts to proxy.ts, and next-intl now reads the locale from next/root-params. Your AI agent still writes the older shape. Four greps tell you which era your setup is from and what to change.
Read article →
Crowdin released an agent skill that wires a JavaScript app to Crowdin in seven phases. We read the source, checked the commit dates and compared it with our own flow. Theirs ends in a Crowdin project with translation switched off. Ours ends in a pull request.
Read article →
v0 hands you a real Next.js codebase, and it already commits to a GitHub branch. That makes the repo-native route open from the first prompt: translation catalogs as committed files, nothing third-party in the render path.
Read article →
You wrote the instructions file and Copilot still hands you hardcoded English. It is not ignoring you — inline completions never read it. Which Copilot surfaces obey your rules, and how to set up i18n around the one that does not.
Read article →
The agent added your loading state and also renamed the button. Why agents treat copy as code they may clean up, why rules files only reduce the frequency, and the workflow that makes every reword reviewable.
Read article →
next-intl for the Next.js App Router, react-i18next for ecosystem breadth, Lingui for compile-time extraction and PO catalogs. How to pick a runtime i18n library in 2026 — and what none of them does for you.
Read article →
Copy drift is user-visible text diverging from its approved source across surfaces and over time, with no single source of truth. What it is, what it is not, and how to prevent it.
Read article →
Missing keys, orphaned keys and locale-file merge conflicts are routing failures, not translation failures. Why the drift happens and what removes it.
Read article →
Copy drift is user-visible text diverging from its approved source with no single source of truth. What causes it, and why consistency and localization are one problem in a codebase.
Read article →
Keep the whole translation workflow in your repo: committed catalogs, a locale config, and a compile step in your build. No login, nothing your coding agent cannot reach.
Read article →
AI localization automates the whole workflow, not just the translating: string extraction, translation keys, locale files, and sync on every push.
Read article →
The seven steps to localize a website, the four-layer tool stack, what it costs, and how a coding agent now does the setup work a team used to do.
Read article →
Seats, hosted or processed words, and tier gates each grow on their own schedule. Here is how the three multipliers compound, and what removing two of them changes.
Read article →
Two ways to translate a Lovable app: a runtime widget or locale files committed to your repo. Steps for both, and which survives a vendor shutdown.
Read article →
Commit Lingui PO catalogs to your repo and translate on every Git push. No runtime widget, no vendor lock-in, and translated HTML rendered server-side.
Read article →
Your Lovable app translates into Arabic but the sidebar stays left and the chevrons point the wrong way. Fix RTL in your repo, not a widget.
Read article →
Cursor re-adds hardcoded strings every time it regenerates a component. Here is why it happens and how to catch and fix them on every push.
Read article →
Google ranks pages by their visible language. What localizing your app actually does for SEO and how to make translated pages rank.
Read article →
The real shift in localization is position, not translation quality: out of the export-translate-import round-trip and into the dev loop, shipping as a localized PR.
Read article →
Lovalingo closes August 31, 2026. What breaks in your Lovable app, how to test it in DevTools today, and how to move translations into your own repo first.
Read article →
Your Lovable app's UI is translated but Supabase rows still render in English. Localize database content in your own schema, no runtime widget.
Read article →
Lovalingo closes Aug 31. How to export your translation bundles, convert them to PO catalogs in your repo, and never lose translations again.
Read article →
Locize starts at $7 and LocaFlow at $9, but the sticker price is the least interesting number on a localization pricing page. Here is what thirteen tools charge at entry, what each one meters, and what that means at 10,000 keys across 5 locales.
Read article →Self-hosted Tolgee is the only route with no licence fee and no key limit. Every hosted free tier stops somewhere small, and the do-it-yourself route bills you in evenings. Here is what the zero-cash options really cost.
Read article →Lovalingo closes on August 31, 2026, and runtime translations disappear with the service. Compare the realistic alternatives for Lovable apps, from committed catalogs you own to runtime widgets that rent your languages back to you.
Read article →
Your Lovable app is translated, the switcher works, and Google still shows only English. Give every language its own URL, wire the switcher to navigate, and emit reciprocal hreflang so each locale can rank.
Read article →
Most Lovable apps run the classic Vite plus React SPA stack, the one the recent i18n tooling wave skipped. Here is how to add i18n with committed Lingui PO catalogs that sync on every push, with no runtime widget and no vendor lock-in.
Read article →
Lovalingo closes on August 31, 2026, and its runtime translations shut off with it. What breaks, what to export, how to remove it cleanly, and how to rebuild your languages as files in your own repo — with three months of usage free for migrators.
Read article →
The best way to localize an app Claude Code generated is to stay in the agent: one command installs a localization skill, Claude Code wires the i18n, and translations arrive as pull requests on every push. The full loop, from English-only repo to merged multilingual PR.
Read article →
SEO translation adapts your pages, keywords, and metadata so translated pages actually rank, not just read correctly. Here is how it differs from plain translation, the technical layer search engines require, and the workflow that survives constant shipping.
Read article →
An AI localization agent keeps your app translated inside your coding workflow, not in a separate dashboard. Here is what it does, how it differs from machine translation, and what infrastructure it needs to be reliable.
Read article →
The globalize.now MCP server lets Claude Code, Cursor, and any MCP client run localization from your editor: create projects, translate locale files, and open Git translation PRs. It does the backend; the skill still internationalizes your code.
Read article →
An AI agent can translate text, but it cannot make an app multilingual on its own. Here is the agent-native way to ship a multilingual app from Cursor, Claude Code, or Lovable, and where the infrastructure layer fits.
Read article →
New Lovable projects ship on TanStack Start with SSR, which breaks the runtime translation widgets most guides still recommend: they flash English and get indexed in the wrong language. Committed Lingui PO catalogs render each locale on the server and stay in sync on every push.
Read article →
Software localization adapts an app for a specific language, region, and market, well beyond translation. Here is what it includes, how it differs from internationalization, and how to do it without i18n debt in AI-generated apps.
Read article →
Weglot translates a Lovable app at runtime in the browser; globalize.now commits Lingui PO catalogs to your repo and syncs them on every Git push. Here is how the two compare on SEO, pricing, and ownership.
Read article →
Lovalingo and Weglot translate Lovable apps at runtime in the browser; globalize.now commits translated catalogs to your repo and syncs them on every Git push. Here is how the three approaches differ.
Read article →
ICU MessageFormat packs plurals, gender, and variables into one translatable string, but its syntax was built for code, not translators. Here is why it is still hard to localize, what MessageFormat 2.0 changes, and where it breaks in AI-generated apps.
Read article →
Lokalise paid plans start at $144/mo and are priced per seat, with no permanently free tier — only a 14-day trial. For a solo dev keeping locale files in sync, the real bill climbs with seats and word caps — while per-workspace pricing with no per-seat or per-language charges keeps it predictable.
Read article →
Most Lovable multilingual guides stop at the first build. This one shows the durable setup: Lingui PO catalogs, the lovable-i18n skill, and automatic translation sync so your app stays fully translated as you keep shipping.
Read article →
The next localization default is an open agent skill: repo-native instructions your coding agent reads, applies, and leaves as standard locale files, without a new SDK or hosted MCP dependency.
Read article →
Tolgee, Locize, SimpleLocalize, LocaFlow, Prismy, Weglot, POEditor, Crowdin, Lingo.dev and Languine mapped by price and fit, re-verified August 2026. For AI-coded apps with no keys yet, the real gap is upstream string extraction, and globalize.now handles that on every Git push.
Read article →
Agent Skills are spreading across Claude Code, Cursor, and Codex. npx skills add globalize-now/globalize-skills installs the localization Skill. Here is why that slot is becoming the default.
Read article →
If next-intl, i18next, or Lingui is already in your repo, L1 mode skips the destructive scaffold: no routing rewrite, no surprise switcher, just locale PRs on every push. Detection signals, hard boundaries, recovery if the wrong branch runs.
Read article →
Install one CLI, send Cursor one prompt, and turn a hardcoded English Next.js App Router project into a multilingual app with next-intl keys, locale files, and Git-push translation sync.
Read article →
We pointed our AI agent at a production monorepo with Next.js, Android, and MCP servers mixed together. Here is exactly what happened: 28 UI string candidates extracted, 41 entries converted, 5 languages added, and production-quality translations confirmed.
Read article →
Lovable prompts keep rewriting UI with new hardcoded strings while PO catalogs fall behind. Here is why that gap widens every release and how Git-native i18n sync closes it without abandoning your builder.
Read article →We audited our own five-locale marketing JSON. Cyrillic in German, formal Sie and vous on a dev tool, invented euros, encoding false friends, and shipping verbs that meant parcels. Then we wrote policy and a verify script so it regresses less quietly.
Read article →
Install agent skills, run five prompts, and go from hardcoded English to synced locales and a language switcher—without weeks of manual i18n work.
Read article →
English reaches a fraction of the world's online users, yet half the web stays English-only. Here's the conversion data, the ROI case, and why AI-built apps make 'later' more expensive than ever.
Read article →
Map the full stack in three layers — i18n architecture, translation management, and execution engines — so you know which tool to reach for first when your app is still mostly hardcoded English.
Read article →
Lokalise and Crowdin shine at translation workflows — once your app already speaks in keys and locale files. Most AI-built products skip that step. Here is how to think about tooling when your codebase is still the wild west.
Read article →
You do not need to throw away an AI-built codebase to support new languages. Start with i18n architecture, extract strings into keys and locale files, then layer glossaries and automated translation with real QA.
Read article →
Localization didn't start with AI, It didn't even start with TMS's, It started decades ago with tools built for translators, not product teams. And that legacy still shapes how the industry operates today.
Read article →
Traditional localization tools assume your app is already internationalized. Most modern AI-generated apps are not.
Read article →