Small builds, real experiments

Why we do this in public

Labs is where we find out what AI is good for, by building something real with it and letting people use it. It keeps our advice honest: we don't recommend an approach to a client that we haven't run into the ground ourselves first.

It's also the proof. Anyone can describe what AI might do. A lab you can open in a browser is harder to argue with.

The hard part is finding work that splits cleanly. When a project has parts that can run separately without tripping over each other, we run them at once. Often it doesn't split, and then it's one thing at a time.

Guten Glance

live at gutenglance.com

Lab 01 · live

Discover gems from classic texts

Project Gutenberg holds tens of thousands of books, all free to read. The hard part is knowing where to start. Guten Glance picks a book, finds the passage worth reading, and puts it in front of you.

How it uses AI

Every day it picks a book at random from Project Gutenberg and parses it. Gemini 3 Flash reads the book and picks the passage most worth reading, under one firm instruction: don't give anything away. It then tags the excerpt by theme, so a passage about a boy choosing the wrong toy files itself under conduct-of-life, foresight, growing-up.

That constraint is the interesting part. "Find the best passage" is a search problem. "Find the best passage that doesn't spoil the book" asks the model to understand what the book is doing. That is the kind of judgment people assume AI can't be trusted with. It can, if you ask properly.

That's the whole thesis of this company, in a package small enough to test in a week. We designed and built the site AI-natively, the same way we build for clients.

• AI selection & taggingGemini 3 Flash picks a spoiler-free excerpt from a randomly chosen book and tags it by theme.
• Read the whole thing, freeEvery passage links straight to the full book on Project Gutenberg. It's public domain, so there's no paywall and no account.
• Social layerProfiles, bookmarks, comments, replies, and up/down voting, so a passage can start a conversation.
• Run properlyA full admin and moderation system behind it. The refresh schedule is configurable.

Simul

in build · Chrome extension

Simultaneous translation for the web. Simul puts a page in two languages side by side: the real page on one side, a read-only mirror of it in your language on the other. You work in the real page, and you read the mirror.

Browser autotranslation rewrites the page you are using. On an ordinary article that is fine. On a booking form, a government site or an order screen it is not: swapping the text out from under a form is what breaks the form. Simul never touches the page you are filling in. The translated copy sits beside it and is inert by design, so nothing can be clicked or submitted there by mistake.

It runs in Chrome, and the translation happens on your own machine. The source is on GitHub under the MIT license, and we will write up how it is built when it ships.

What's next

More labs are in progress. If one proves useful, it graduates into a real product; if it doesn't, we learned something cheaply and moved on. Either way, the week was worth it. Feedback is always welcome at hello@pococo.ai.