When I built Talk to Sammie, I wanted people outside the United States to be able to use it.

By July 2026, Sammie had been checked for natural conversation in English, French, German, Italian and

Spanish.

That felt like a significant accomplishment.

My conversational AI was multilingual.

Then I began discovering that a multilingual AI and a multilingual product are not the same thing.

Sammie could speak the language

The conversational part worked surprisingly well.

Sammie could respond naturally in French, German, Italian and Spanish, and I had also paid attention to

differences within English, such as American and British usage.

But the conversation was only one part of the customer's experience.

The buttons were still in English.

The surrounding interface was still primarily English.

A person could potentially have a perfectly reasonable conversation with Sammie in Italian while interacting

with an application around that conversation that was still speaking English.

Technically, I had multilingual capability.

From the user's perspective, I had something much less complete.

International visitors made the distinction matter

This wasn't originally an abstract localization project.

I started paying more attention because people outside the United States were actually showing up.

That led me to translate landing pages into French, German, Italian and Spanish and localize things such as

currency.

Later, the site evolved again.

Instead of automatically forcing someone into another language, the main landing page remained English.

A browser using Italian, French, German or Spanish could trigger a banner offering the appropriate

localized version, and visitors could also select a language themselves.

That was another distinction I hadn't thought much about when I started.

Supporting a language wasn't the same as deciding when the product should use it.

Then people arrived speaking languages I hadn't planned for

Traffic also started appearing from places such as Portugal and Brazil.

Portuguese wasn't one of the five languages I had deliberately optimized.

That created another question I hadn't anticipated:

Could the underlying AI probably carry on a basic conversation in Portuguese anyway?

Probably.

But that didn't mean I should suddenly claim that Talk to Sammie was a Portuguese-language product.

I decided not to keep adding languages simply because the underlying model might be capable of them.

Unless actual customer demand justified the work, five supported languages were enough.

The capability was the easy part

This was one of the stranger things about building with AI.

The technically impressive part—having an AI capable of conversing in several languages—was surprisingly

accessible.

The difficult questions appeared around it.

What language should the interface use?

Should the site switch automatically?

Should it ask?

What happens to buttons and navigation?

What currency should someone see?

How much testing is enough before I claim that a language is actually supported?

What happens when the underlying AI can speak a language that the rest of the product doesn't officially

support?

Those aren't really translation questions.

They're product decisions.

What changed for me

I stopped thinking of “multilingual” as a feature that could simply be turned on.

There were layers:

The AI can understand the language.

The AI can respond naturally in the language.

The interface supports the language.

The marketing and landing pages support the language.

The business is prepared to call that language officially supported.

Those layers don't automatically arrive together.

That seems obvious after building it.

It wasn't obvious before I did.

The surprising part wasn't that localization required work.

The surprising part was that AI made the hardest-looking piece—actually having a conversation in another

language—so easy that it exposed all the less glamorous pieces surrounding it.

My AI spoke five languages before my product did.

Skip the expensive detours.  Here's the roadmap I wish I had -https://secondlookworks.com/products/before-you-build-it-a-realistic-roadmap-for-starting-a-business-without-wasting-time-and-money-pdf-download