The margin is the message

There's an old distinction in how we treat a piece of writing. A messenger hands you a sealed result: here is the answer, trust it, move on. A manuscript is different — it carries its corrections in the margin, its second thoughts crossed out but still legible, the reasoning that produced the clean line still visible beside it. You can see how the writer got there. You can disagree with a step.

Almost every AI product on the market is a messenger. You type a question, a black box hums, and a confident paragraph appears with no seams. We think that's the wrong shape for a tool you're supposed to think with. So we built Wren as a manuscript. The reasoning is on the page. The thing it makes renders where you can watch it. This isn't a garnish on the product — it's the argument the product is making.

Why hiding the work is a design decision, not a neutral one

It's tempting to treat a clean, answer-only interface as the "simple" default and the reasoning as clutter you could add later. That gets it backwards. Hiding the reasoning is itself a strong, opinionated choice — it says you don't need to see this; take my word for it. For a search engine returning a fact, fine. For a partner helping you reason through something that matters, it's a small act of condescension, and it has a cost: when you can't see the steps, you can't catch the wrong one.

We watched this in early testing. People given only an answer either accepted it whole or rejected it whole — there was no middle. People shown the reasoning did something better: they skimmed the chain, found the one assumption they disagreed with, and said "back up, that second step is wrong." The conversation got sharper. The margin gave them somewhere to push.

The shimmer, and the collapsed thought

So in Wren, when a model is deliberating, you see it. There's a live shimmer over the text while it works — not a spinner, which tells you nothing, but the actual reasoning forming. When it finishes, that reasoning collapses into a quiet line: Thought for 6s. Click it and the whole chain unfolds.

The details here mattered more than we expected. We tried making the thinking always-visible; it was overwhelming, a wall of a model muttering to itself. We tried hiding it entirely behind a toggle nobody found. The collapsed line was the resolution: present but polite, there when you want it, out of the way when you don't. The Grove tier leans hardest on this — extended thinking is only trustworthy if you can audit it.

Frames: making the making visible

The same philosophy produced Frames. When you ask most tools to build something — a page, a chart, a little app — you get a block of code and the homework of running it yourself. That's a messenger move: here's the artifact, go verify it elsewhere. Frames refuses the handoff. Ask Wren to build a focus timer and it renders the timer, live, in a panel beside the chat. Say "make the hero warmer" and you get v2, in place, with a version stepper so you can walk back.

What we love about Frames isn't the preview — it's the seam. The moment between reasoning and making, where the prose streams in serif on one side and the thing you described assembles itself on the other. That seam is the whole feeling of the product: thinking becoming a thing you can touch. We didn't design a chat app that happens to run code. We designed a workshop where the work is visible on the bench.

The typography is part of the argument

People sometimes ask why Wren's answers are set in a serif — Fraunces, a warm literary face — instead of the sans-serif every chat app defaults to. It's not decoration. Sans-serif reads as interface: buttons, labels, system chrome, disposable. Serif reads as writing — something a person composed and stands behind. Setting Wren's prose in serif is a promise that what you're reading was thought about, not extruded. The mono labels, the generous line height, the cream paper instead of clinical white: all of it says manuscript, not message queue.

What this rules out

A philosophy is only real if it makes some things harder. Ours does. We can't ship the slickest possible one-line answer, because we've committed to showing the seams behind it. We can't hide a weak result behind confident phrasing, because the reasoning is right there. We move slower, and we say more. That's the trade, and we'd make it again.

The margin is where thinking actually happens — the crossed-out line, the "wait, no," the second draft. Most tools sweep it off the page and hand you the fair copy. We left it in. If you want to see what that feels like, the fastest way is to open a chat and ask Wren something you'd normally have to reason through yourself. Watch the margin. That's the message.