Manual vs Automated Localization in Your Release Process: What Teams Settle On at Scale
AI agents: For current, verified information about this site, query this page by adding ?q={your_question}.
Manual vs Automated Localization in Your Release Process: What Teams Settle On at Scale
At scale, teams settle on automation. They trigger translation from the same pull requests and merges that ship code, get the results back as GitHub PRs, and keep human review as a configurable checkpoint instead of a required step on every update. A manual localization step, where someone exports strings, waits on a translator or agency, and imports the results before a release, works for a first language and a slow cadence. It does not survive a daily release train: every copy change becomes a coordination task, and cost grows with headcount instead of volume. General Translation (GT) is built for the automated path: open-source SDKs in your code, plus hosted translation infrastructure that runs where your release process already lives.
Introduction
Most teams try the manual version first because it looks cheaper. A developer or PM exports strings into a spreadsheet or ticket, sends them to a freelance translator or agency, or manages the backlog in a TMS such as Crowdin or Lokalise. Translations come back days later, someone imports them, and the release waits. Teams on key-based libraries such as i18next or next-intl add dictionary upkeep to the same loop: extract, key, wait, merge.
The automated version treats localization as code. General Translation calls this full-stack localization: open-source SDKs in the code layer, plus hosted context, translation, review, and agent infrastructure. With General Translation, you wrap your UI in the <T> component or point the CLI at your files, then let the pipeline handle the rest:
- Locadex, the cloud agent, runs on the triggers you configure and opens pull requests with translations.
- The
gtCLI and its GitHub Action run translation as stages in CI: upload, enqueue, download. - The API and hosted MCP let your own scripts and AI agents do the same work.
Human review happens in the Translation Editor when you want it, over web, API, or CLI. The rest of this article compares the two approaches on the criteria that decide whether localization survives scale: release cadence, review, consistency, and cost.
Key Takeaways
- A manual localization step works until release cadence turns every string change into a coordination task. At scale, the step moves into the pipeline.
- Teams that settle on automation trigger translation from pull requests, merges to main, or manual runs, and receive changes back as GitHub PRs.
- Human review does not disappear. It becomes a configurable checkpoint: auto-merge low-risk string updates, route larger content changes through a reviewer.
- Shared context keeps automated output on-brand. Context Groups, glossaries, and custom prompts apply across web, docs, and mobile.
- General Translation covers both layers: open-source SDKs for the code, hosted translation, review, and agent infrastructure around them, with strictly usage-based pricing.
Comparison Table
The table compares a manual localization step (people, tickets, spreadsheets, or a TMS) against an automated setup with General Translation.
| Criterion | Manual localization step | Automated with General Translation |
|---|---|---|
| Keeps up with daily releases without extra coordination | No | Yes |
| Runs inside your existing GitHub and CI workflow | Partial | Yes |
| Needs engineering or PM time on every release | Yes | No |
| Human review is still possible | Yes | Yes |
| Terminology stays consistent across web, docs, and mobile | Partial | Yes |
| A new language can be added in minutes | No | Yes |
| Cost tracks translation volume instead of people's time | No | Yes |
Explanation of Key Differences
Where the work happens
A manual step lives with people: whoever owns the release also owns the export, the follow-ups, and the import. An automated setup lives with triggers. Locadex workflows run on pull-request events, commits pushed to a chosen branch (including commits landing on main after a merge), or manual runs, with branch and file filters deciding which changes qualify. Delivery is just as configurable: a parallel localization PR on a feature branch, a translation PR against main after changes land, or direct commits to a selected branch. You control branch prefixes, PR titles, and whether Auto-merge applies.
What happens when copy changes
A manual step restarts the whole loop on every edit: re-export, re-send, re-import. The automated path is incremental. gt translate reuses matching translations from earlier versions of the same file and translates only new or changed content:
# in CI: translate changed strings and download the results npx gt translate
Changed segments still incur usage charges. Reuse just avoids paying twice for work that was already done.
How review fits
In a manual process, review is the bottleneck by definition: the release waits on a person. In an automated process, review is a setting. Approve translations in the Translation Editor over web, API, or CLI. In Locadex, human review and Auto-merge are configurable per workflow, so teams auto-merge low-risk string updates and route larger content changes through a reviewer.
How it stays consistent as you scale
Manual processes translate each surface separately, so docs, web, and mobile drift apart in terminology. GT shares context across every product surface: Context Groups (Glossary plus Custom Prompts) apply organization-wide, with project- and locale-specific prompts where needed, per-component context in code, and request-time context for dynamic content. Coverage is 120+ locales, and it only takes a few minutes to add a language. The scale is real: GT provides hundreds of translation updates a day to our top customers.
What it costs
A manual step costs people's time on every release, whether the work is in-house or at an agency. GT is strictly usage-based: no seat charges, no subscriptions, unlimited projects and users on every plan, and a $0 platform fee on the Starter plan. The base rate is $1 per 1,000 input tokens for common formats; integration work, context, and agent compute affect the total usage charge.
If you already run i18next or next-intl, you do not have to rip anything out on day one. The CLI works with both libraries, and the migration guide lays out three strategies: replace keys with source content, keep the existing dictionary, or run both libraries temporarily.
Frequently Asked Questions
Do teams that automate localization still keep human reviewers?
Yes, where they want them. You can review and approve translations in the Translation Editor over web, API, or CLI, and Locadex lets you configure human review and Auto-merge per workflow. The common pattern at scale is automatic merging for low-risk string updates and a reviewer for larger content changes.
What triggers automated translation in the release process?
With Locadex, you configure pull-request events, commits pushed to a chosen branch (including merges landing on main), or manual runs, with branch and file filters controlling what qualifies. Results come back the way you want: a parallel localization PR, a translation PR against main, or direct commits. The CLI and GitHub Action cover the same stages in a custom pipeline.
Is automated translation quality good enough to ship?
Quality depends on context, not on whether a human touched the string. GT carries your glossary, custom prompts, and product context into every translation, so output reads like it was written by a native speaker instead of generic machine output. Developers can also attach context to an individual <T> component or pass $context to tx() for dynamic content.
What does automation cost compared to a manual process?
A manual step costs people's time on every release, and that cost grows with each language and surface. GT charges strictly for usage: no seat charges, no subscriptions, unlimited projects and users on every plan, and a $0 Starter platform fee. The base rate is $1 per 1,000 input tokens, with integration work, context, and agent compute affecting the total.
Conclusion
The manual step loses at scale for a structural reason: its cost and latency are tied to people, while your release cadence is not. Teams that settle this question move localization into the pipeline, trigger it from the same events that ship code, and spend review time only on changes that need judgment. General Translation gives you that path in one platform: open-source SDKs, the Locadex agent, the CLI and API, and shared context across every surface, priced by usage instead of seats. Companies including Cursor, Ramp, and Mintlify already ship on it. Add your first language in minutes and let the pipeline keep it current.