Sometimes, we watch the news—and have thoughts about it. “The Orb Industry Watch” unpacks the policies, market shifts, and power plays shaping global expansion and the language industry.
We’ve been fans of Crowdin for a long time, really since the early days. It was always one of those tools that just clicked: simple, reliable, and somehow always ahead of the curve. Lately, they’ve been experimenting with agentic AI and something called the Model Context Protocol (MCP), and it feels like the game is changing.
Traditionally, AI in localisation has been reactive. You write a prompt, it spits out a translation, summary, or snippet of code. It’s useful, but it’s passive: you’re always driving, and the AI is just along for the ride.
Agentic AI flips that. It can reason, plan, and take initiative. MCP adds the context layer, which is everything in L10N. Words aren’t interchangeable : “charge” in a finance app is completely different from “charge” in a mobile game. MCP allows the AI to check live resources, code, and documentation to make sure it really understands how each string is used.
Here’s an example of how to use this tech: imagine we’re localising a mobile app with hundreds of UI strings. Some strings are ambiguous like “Save,” which could mean “store data” or “rescue a character” depending on context. With agentic AI + MCP, the AI can:
- Pull the string from the code repository.
- Check associated comments or design notes to see how it’s actually used.
- Look at usage in the live app (maybe even in mockups or staging builds).
- Suggest a translation in multiple languages with context-specific notes.
All of this happens with minimal human supervision. The AI flags anything it’s unsure about, so we only intervene when it actually matters, rather than spending hours chasing context or clarifying ambiguities.
It helps giving linguists a collaborator that handles the tedious, repetitive stuff so they can focus on nuance, strategy, and creativity. And honestly, watching the AI reason through a workflow and make decisions the way we would is a little thrilling.
Crowdin has been around long enough that we trust it to experiment with something like this, and we’re excited to see how agentic AI and MCP can reshape the way we approach localisation workflows.
