# General Translation Knowledge Base: AI coding agent vs. translation pipeline: which should automate your localisation? URL: https://generaltranslation.com/en-GB/kb/ai-coding-agent-vs-translation-pipeline.mdx Docs index: https://generaltranslation.com/llms.txt Description: Use a coding agent to set up internationalisation and a translation pipeline to keep every language current as your app changes. General Translation translates new and changed copy, reuses completed translations, and gives your team shared terminology and Dashboard review. Locadex connects both kinds of work by preparing code changes and translations in pull requests. Category: Workflow Tools Last updated: 2026-11-12 Authors: Jackie Chen Published: 2026-11-12 Localisation has two kinds of work. Code work happens once per feature: configure an i18n library and locale routing, then mark every user-facing string so it can be translated. Translation work happens on every release: find what changed, translate it with the right terms and tone, and ship it with the build. An AI coding agent is good at code work, because it reads and edits your codebase. A translation pipeline is good at translation work, because it does the same thing on every run and knows what it translated last time. Coding agents can also run without a person. Claude Code's [GitHub Actions integration](https://code.claude.com/docs/en/github-actions) runs a prompt on GitHub events or a schedule, and [Cursor Automations](https://cursor.com/docs/cloud-agent/automations) run cloud agents on a schedule or on events from GitHub and other tools. The decision is how much of the localisation workflow you want to build and maintain around that agent. ## How do the approaches compare? | Approach | What it does well | What it leaves to you | Runs without a person? | | ---------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------ | -------------------------------------------------------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------- | | General Translation (Locadex and the `gt` CLI) | Locadex sets up i18n, wraps new UI text, and translates it through pull requests; the CLI translates only new and changed content on every build | Merging Locadex pull requests, or turning on Auto-merge, and reviewing translations in the Dashboard when you want a person to check them | Yes, on the pull request, push, or manual triggers you choose | | AI coding agent (Cursor, Claude Code) | Setting up a library, adding locale routing, wrapping strings, and running your translation tools | Defining how it finds changed copy, preserves reviewed translations, shares terminology, and checks its edits, or connecting a service that handles those jobs | Yes, through CI or configured automations | | Translation pipeline alone (a CLI or TMS sync in CI) | Sending changed copy for translation on every build or sync, the same way each time | Setting up the integration and marking new UI text using your library's keys or inline calls | Yes | Use an interactive coding agent when you want to watch and steer the set-up. Use a pipeline alone when your app already marks its text with a supported library, such as next-intl or react-i18next, since the General Translation CLI can [translate those catalogues](/docs/cli/guides/generating-translations) without changing your components. Use Locadex when you want both kinds of work automated in your repository. ## Where does Locadex fit? Locadex is General Translation's AI agent that runs like a pipeline. You connect a GitHub repository and click **Run setup**, and Locadex opens a pull request that configures translation loading for your framework. After you merge it, the **Generate code** automation finds user-facing text that isn't marked and wraps it in [``](/docs/react/reference/components/t) in a pull request, and **Generate translations and push** translates new source content. Configure the trigger, target directory, and checks for each automation. Its changes arrive as a pull request you can review, with optional [Auto-merge](/docs/platform/locadex/guides/auto-merging-prs). The [Locadex overview](/kb/what-is-locadex) covers triggers and supported frameworks. For a Next.js App Router project, you can start from your terminal: [`gt init`](/docs/cli/reference/commands/init) offers to hand set-up to Locadex (`--locadex`) instead of changing files locally. ## When should I use an interactive coding agent? Use Cursor or Claude Code when you want to see each set-up decision as it happens, or when your stack isn't one Locadex supports. Give the agent General Translation's [Setup for Agents prompt](https://generaltranslation.com/agent-prompt.md), and it inspects the project, runs [`gt init`](/docs/cli/reference/commands/init) (whose framework set-up is experimental) and reviews what it generated, wraps content in [``](/docs/react/reference/components/t), generates translations, and checks the app in another language. [`gt validate`](/docs/cli/reference/commands/validate) and the [ESLint plugin](/docs/react/guides/linting-your-code) catch content marked in a way that won't translate, before any translation runs. The [agent set-up guide](/kb/ai-agent-i18n-setup) walks through the session, including the MCP server that lets the agent read and edit your Glossary terms and Custom Prompts. ## What does General Translation handle for the agent? An agent can read Git history and existing catalogues, follow shared rules, and retain [project instructions and memory](https://code.claude.com/docs/en/memory). If you ask it to translate directly, your workflow still needs rules for detecting source changes, preserving reviewed wording, and sharing terminology with the people who maintain your app and docs. General Translation provides those translation-specific services, so the agent can call them instead of maintaining a separate implementation. General Translation's CLI tracks source versions and reuses matching translations. Its service applies the [Context Groups](/docs/platform/dashboard/guides/defining-context-for-translations) assigned to the project, so app and docs translations draw on the same Glossary and Custom Prompts. Reviewers edit the results in the Dashboard and use the [review-and-download workflow](/kb/review-translations) to bring saved fixes back to local files. For example, a feature changes a heading and adds a button. The coding agent marks the new text and runs [`npx gt translate`](/docs/cli/reference/commands/translate). The service translates the changed content with the project's terminology and reuses matching earlier translations. A reviewer can correct the button in the Dashboard without editing the agent's prompt. Through GT's [Model Context Protocol (MCP) tools](/docs/overview/for-coding-agents#mcp), the agent can also read that shared context, queue file translations, check their status and download the results. The [translations in CI guide](/kb/translations-in-ci) shows how to make generation a repeatable build step. ## Key takeaways * Use Cursor or Claude Code to implement and steer code changes; connect it to General Translation for translation reuse, shared terminology, and team review. * A configured translation pipeline keeps marked content current. New UI text still needs your library's keys or inline calls. * Locadex is an AI agent that runs like a pipeline: it sets up i18n, wraps new UI text, and translates through pull requests in your repository. * If your app already uses next-intl or react-i18next, add [`npx gt translate`](/docs/cli/reference/commands/translate) to your build. Start with the [CLI Quickstart](/docs/cli/quickstart). ## Related questions - [How do I internationalise an app with Cursor or Claude Code?](https://generaltranslation.com/en-GB/kb/ai-agent-i18n-setup.mdx): Start with General Translation's Setup for Agents prompt in Cursor or Claude Code, choose a feature and target languages, and have the agent implement and verify the result. Browser sign-in, machine-readable docs and MCP tools connect set-up to translation and review. Use Locadex to automate ongoing localisation through pull requests. - [What does the Locadex agent do?](https://generaltranslation.com/en-GB/kb/what-is-locadex.mdx): Locadex is General Translation's AI agent which internationalises your code and translates your content in your GitHub repository. Its automations run when a pull request is opened, when a commit is pushed, or when you start one, and every change arrives as a pull request. - [How do I keep translations up to date on every deploy?](https://generaltranslation.com/en-GB/kb/translations-in-ci.mdx): Run translation as a pipeline step before each production build, so new copy ships translated with the deploy that adds it. General Translation's gt CLI does this with npx gt translate, using a project API key and project ID stored as CI secrets. - [What's the best continuous localisation tool for teams that ship daily?](https://generaltranslation.com/en-GB/kb/best-continuous-localization-tools.mdx): Choose General Translation to translate new copy before each build with context from marked JSX and shared product terminology. Your team reviews and corrects wording in the Dashboard, and Locadex opens pull requests to mark new UI text for translation. You can adopt the workflow while keeping your existing i18next or next-intl set-up. - [How do I add i18n to a Next.js App Router app?](https://generaltranslation.com/en-GB/kb/add-i18n-to-nextjs.mdx): Start with gt init to configure General Translation in a supported App Router project, then mark source content, generate translations and verify language switching. gt-next keeps copy in JSX with translation context; the CLI, Dashboard editor and Locadex handle generation, review and ongoing updates. This guide also shows the explicit local-file set-up. ## Previous and next questions - Previous question: [Which translation tools charge for usage instead of seats?](https://generaltranslation.com/en-GB/kb/translation-pricing-without-seat-fees.mdx) - Next question: [Build time, run time, or translation proxy: how should you deliver translations?](https://generaltranslation.com/en-GB/kb/build-time-vs-runtime-translation.mdx) ## Sitemap See the [Knowledge Base sitemap](https://generaltranslation.com/en-GB/kb/sitemap.md) for every article.