Internationalize your codebase in under an hour with the Kilo Code AI Agent

Step by step: how we used the Kilo Code AI agent to extract strings, wire up i18n, and ship a fully internationalized React codebase in under an hour.

At Prismy, we used to see internationalization as a massive headache: tedious, time-consuming, and with little immediate business value. But that was before the rise of AI agents like Kilo Code.

The classic problem: hitting the i18n wall

Early-stage teams move fast. User-facing texts are hardcoded everywhere. But the day inevitably comes when you need to support a second language. And that's when the fear kicks in:

hundreds of files to modify,
risk of git conflicts,
weeks of manual work.

Some teams keep postponing. Others hire freelancers. But today, there's a much faster, scalable option.

AI agents like Kilo Code change the game

Where a human developer would painfully hunt for hardcoded strings, an AI agent like Kilo Code excels:

identifying user-facing texts,
cleanly extracting them into translation files with well-organized namespaces,
continuously ensuring the codebase stays functional.

We tested Kilo Code on a reasonably large React project: 286 files, including 121 with hardcoded text. The result? Complete internationalization in under one hour.

🛠️ Our process

Our starting stack

Frontend: React
Backend: Node.js
Goal: extract all hardcoded English strings
AI tool: Kilo Code (VS Code extension)
Translation management: Prismy

1. Installing Kilo Code

Install the extension in VS Code
Provide the OpenAI API key
Select the orchestrator agent (which coordinates the code, debug, and architect sub-agents)

2. The prompt we used

Here is the prompt we provided to Kilo Code to launch the mission:

We need to internationalise our codebase, currently hardcoded in English.
What you need to do:
- knowing that our stack is JS/TS ; react on the front and node in the backend, choose i18n libraries compatible with both front & back, at least with the same syntax
- setup the lib and put a language selector in the user setting (do not need to be persisted in the backend for now)
- browse all the files of our codebase, and identify user-facing hardcoded texts, use the i18n library
- make sure to smartly use the i18n namespace features, so the keys are well sorted in the file
- for now, only create the key/values for the english file, we will later fill the french, do not translate anything for now
- make sure the project builds and run

3. The AI agent at work

Within minutes:

Kilo Code installed i18next (smart choice, even though we hadn't specified any library),
updated the configuration,
started extracting hardcoded strings,
ensured the app still compiled correctly after each change.

Watching files edit themselves live was truly impressive. After about 5 minutes, the project was partially internationalized and still fully functional.

4. Real-time collaboration

During the first pass, a namespace config issue caused keys to appear in the UI instead of translated texts. I simply sent a screenshot via Kilo Code's built-in chat. Within seconds, the agent identified and fixed the problem without me writing any code.

5. Divide and conquer

Instead of running a full extraction at once, I proceeded feature by feature (about 20 files per batch). After roughly ten iterations and less than an hour of actual work, the entire codebase was internationalized.

One PR later: done.