Should you still learn a language in the age of AI?

Real-time translation is genuinely good now. So is it still worth the years it takes to learn a language? The honest case — that a language is a way of thinking, not just translating — and how AI changes the way you should go about it.

Real-time translation is genuinely good now. You can point your phone at a menu, hold a passable conversation through an earbud, and read a foreign news site as if it were written in your own language. So it's a fair question, and people ask it seriously: is it still worth spending years learning a language when a machine can do it for you instantly?

I think it is. But the honest answer isn't "yes, obviously" — it's "yes, for some reasons and not others, and AI should change how you do it."

What translation actually solves

Let's give the machines their due. For practical, everyday use, translation is now good enough — the machines have basically won. Ordering food, finding the right platform, understanding a contract clause, getting the gist of an article — these are all about moving information from one language into another, and that's exactly what the models are good at. If your goal is to not get lost in a city you'll visit once, you do not need to learn the language. Use the tool. It's fine.

If your encounters with another language are mostly practical — ordering, asking, skimming a page — then learning it is a hobby, not a necessity. That's a perfectly good reason to learn — but it's worth being honest that it's the reason, because it changes what "success" looks like.

What translation doesn't touch

Here's where the machine drops out. A language isn't only a protocol for moving information. It's how you think, how you connect, and how you're perceived.

• Connection. People don't open up to an earbud. The moment you speak someone's language — badly, even — the relationship changes. You stop being a tourist to be served and start being a person worth talking to. No latency-free translation replaces that, because the effort itself is the signal.
• Presence. Understanding in real time — catching the joke as it lands, not three seconds later through a translation — is a different experience from having it relayed. If you live somewhere, work with people, or love someone in that language, relayed isn't enough.

None of these are about moving information around. None of them are things you can delegate to a model, because the value is in your having done it. But there's a third reason, and it's the one I find most interesting of all.

A language is also a way of thinking

We talk about languages as if they were interchangeable containers — pour the same thought into English, Polish, or Japanese and out comes the same meaning, just in different packaging. Translation reinforces that illusion, because moving meaning across is exactly what it's built to do. But a language isn't a neutral container. Each one makes different things easy to say — and, over time, easy to think.

I notice it in something mundane. When I write plans and tasks in English, they come out sharper and more concrete than in my first language; English nudges you toward the specific. Another language might pull you toward the formal, the associative, the indirect. None of them is better, but each is a different path of least resistance, and the one you're thinking in quietly shapes the thought itself.

You see it in vocabulary, too. Some languages hand you a dozen shades for the quiet of a night; others give you one plain word and make you earn the rest. Learn the richer one and you start noticing distinctions you previously had no name for. Learn the leaner one and you learn economy. Either way you've added a setting to how you perceive the world — and a genuine door into how the people who speak it see it, which is most of what "understanding another culture" actually means.

This isn't the cartoon version of the idea — that your language is a cage, or that you can't imagine what you have no word for. Decades of research point to something subtler: a language shapes which distinctions are habitual and quick for you, not which are possible. Speakers of languages that map space in fixed compass directions, rather than left and right, keep their bearings in ways the rest of us rarely manage; grammatical gender quietly nudges how people describe the very same object from one language to the next. The effects are real but specific — nudges, not walls. (Guy Deutscher's Through the Language Glass is a good, honest tour of what holds up and what doesn't.)

That's the part that disappears when you reduce language to translation. Picking up another language isn't only gaining a second way to say things — it's gaining a second way to think them. You don't trade in your old defaults; you get a second instrument. A translation app can carry meaning between languages all day, but it can't install that instrument in you. Only learning can.

The part nobody admits: most people quit for a boring reason

If the case for learning is so strong, why do so many people stall? Usually not motivation. It's that traditional learning is full of friction that has nothing to do with the language itself: textbooks pitched at the wrong level, content that's either baby-simple or hopelessly advanced, vocabulary lists with no connection to anything you'd actually say.

For decades, the bottleneck was content — finding material that matched your exact level and interests. That's the part AI quietly fixed. Not the learning. The supply of good material to learn from.

How AI actually changes language learning

This is the shift I find genuinely exciting, and it's the opposite of "let the machine do it for you."

AI removes the content bottleneck. You can now generate unlimited reading and listening material calibrated to your precise level — not a textbook's idea of "intermediate," but your intermediate, on topics you care about. That used to be the expensive, scarce part. Now it's free and infinite.

Which means the bottleneck moves. Once content is unlimited, the thing standing between you and fluency is no longer access — it's active practice. The actual work of producing the language: speaking it, hearing it under pressure, reconstructing it from memory. The model can hand you a perfectly level-matched text every time, but it can't do that work for you. That part is still yours, and it always will be.

So the modern answer isn't "AI replaces learning." It's "AI removes the excuses, and leaves you with the part that was always the point."

How we think about it at LSN Lab

This is exactly the bet behind Lingoshade, one of the products we build. It uses AI for the thing AI is good at — generating texts matched to your exact level, A1 to C2 — and then puts the hard, irreplaceable work front and centre through three methods borrowed from interpreter training and language exams:

• Shadowing — repeating speech in real time to build pronunciation and fluency.
• Dictation — writing down what you actually heard, and seeing the gap word by word.
• Spaced repetition — reviewing vocabulary kept tied to the sentence it came from.

The AI carries the content. You do the practice. That division of labour is, I think, the right way to learn a language in 2026. If it sounds like how you'd want to study, you can try it on lingoshade.com.

So — should you?

If you only ever need to get the gist, no. Let the machine translate, and spend your time elsewhere.

But if you want to connect with people, change how you think, or actually be present in a language rather than have it relayed to you — then yes, still learn it. AI hasn't made that obsolete. It's made it more achievable than it has ever been. The only thing it can't do is the practice. Fortunately, that was always the good part.