---
title: Iterate a slide
description: The visual feedback loop — tweak, comment, apply.
---

Once a deck exists, iteration is the inner loop:

```text
present  →  click to comment  →  /apply-comments  →  repeat
```

Most decks live in this loop. You're rarely typing CSS by hand — you're
running the deck, spotting what's off, and pointing at it.

## Run the deck

Start the dev server and open the slide you want to work on. Use it the way
you'd present it: arrow keys, fullscreen, the whole thing. Iteration is
faster when you're reacting to the deck, not staring at code.

## Tweak directly

For changes that are faster to perform than to describe — re-wording a
headline, nudging a colour, swapping an image — open the inspector
(`i`), click the element, and edit it in place. All edits buffer until you
press **Save**, so a batch lands as a single change.

The [Inspector](/docs/core-feature/inspector) page covers the full panel.

## Comment when it's easier to say than do

When the change is easier to describe than to perform — *"use the accent
colour on this title"*, *"make this whole row breathe more"*, *"this needs a
better visual"* — click the element, pick **Comment**, and type the note.

You can drop a handful of comments through a single rehearsal pass without
breaking flow.

## Apply with one command

When you're ready, ask your agent:

```text
/apply-comments
```

The agent reads each comment, edits exactly that region, and clears the
note. Refresh the dev server and you're looking at the next draft. If a
comment is ambiguous, the agent leaves it alone and tells you why — refine
the wording and run again.

## Talk to the agent about what you're seeing

You can also just ask. *"Shrink this title"*, *"swap the cover image"*,
*"this slide feels cramped"* — the agent already knows which slide you're
on and which element you've picked. No need to say *"slide 3, the H1 on
the left"*.

Combine the two modes freely: tweak what's obvious, comment what's not,
and let the agent close the gap.