Emojis Are All You Need

I’ve always said that even the best engineering team, if trapped in a bubble of its own ideas, can build amazing systems but still miss the features users actually want.

That happened to us right after launch. We focused on model routing, output stabilization, context handling, and deterministic workflows. Meanwhile, users were asking one simple thing:

“Great tool, but why is everything plain text?”

Pendless with a readable summary of the CNN homepage, including links

We had avoided Markdown at first. It lets you add headings, lists, emphasis, and even emojis, but the model kept drifting in and out of it. Sometimes it followed the formatting perfectly, other times it ignored it entirely. With careful prompt design, output checks, and stabilizing behavior across contexts, we finally made it reliable.

And since we were already taming the output, we took the opportunity to add another small but useful feature: the ability to include links in the responses. Now the outputs can point directly to resources, references, or any URL you need, without breaking the formatting.

We just published a new version with these features.

This is exactly what we enjoy doing: listening carefully, adjusting fast, and shipping improvements without ceremony.

P.S.: As a European, I had seen Thanksgiving in movies, but it wasn’t until I moved here that I truly understood it. It’s the most spiritual of holidays, a time to appreciate family, friendship, and love. Wishing everyone a happy Thanksgiving.