Save the Cat

Zach Kazanski · February 27, 2026

Blake Snyder’s screenwriting book Save the Cat names a technique as old as storytelling: early in a story, your protagonist does something small and unrehearsed that makes the audience like them. Not admire. Not respect. Like. The classic example is a hero shooing a cat out of the way of danger before a fight – it costs them nothing and reveals everything.

The technique has a specific structure. You don’t just have the character do the endearing thing. You establish the constraints first. You show the reader what the act would cost. You make the reason not to act visible. Then the character acts anyway, quietly, often without seeming to notice they’ve done anything. The setup is what makes the act land.

I’d been thinking about this in relation to Chapter 1. Andrzej is a hard character to like on first meeting – a Polish soldier who has just sailed across an ocean to suppress a slave revolt he was told was a French rebellion against good order. His arc is one of the longest in the book: eight years of gradual, costly transformation. The reader needs to be on his side from the first pages, before they know who he becomes.

The chapter already has texture. It doesn’t have a Save the Cat moment.

Building the Pass

I wrote a new editing pass for this – pass-save-the-cat. The command takes a file path and a character name. The character name matters because the act has to be calibrated to that specific person: what they notice, what they’re short on, what their body does before their mind decides. An act of decency that fits Andrzej (spare, physical, unremarked) would be wrong for Grotowski (performed, ironic) or Erzulie (deliberate, never accidentally kind).

The pass runs in two phases.

Phase 1 is research and proposal. The pass reads the chapter, reads the character file, identifies whether a Save the Cat moment already exists and how strong it is, then generates two or three specific options. Each option includes: what the act is, the 2-4 sentences of setup that would precede it, where exactly in the chapter it slots, and why it fits this character in particular. Then it stops and presents these to me before touching anything.

This matters more than it might seem. Save the Cat moments are easy to get wrong. If the act is too large it feels plotted. If the character announces what they’re doing it becomes performance. If the setup doesn’t establish cost, the act feels costless, which means it doesn’t reveal anything. The proposals need to be evaluated before any prose is changed, because a bad Save the Cat moment is worse than none – it makes the protagonist feel engineered.

Phase 2 is execution. Once I’ve chosen an option (or rejected all of them and said what I want instead), the pass inserts the setup and the act using the Edit tool. Maximum 3 sentences for the act itself. No named emotions. No internal justification – the character doesn’t explain to themselves why they’re doing this. The act is complete before the narrative comments on it.

What It Looks Like in Practice

The pass audited Chapter 1 and came back with three options. The one I’m leaning toward: Andrzej has a small amount of food – a piece of bread he’s been holding since before they disembarked. A woman near him on the dock is counting something under her breath (the pass identified a moment in the chapter where this could slot naturally). He gives her the bread. No eye contact. No acknowledgment. The chapter continues.

Setup: two sentences establishing he hasn’t eaten since the crossing, one sentence introducing the woman with a single specific physical detail. Then the act. Seven sentences total including the act itself.

That’s it. It’s nothing. That’s the point.

Why the Two-Phase Design

Most of our editing passes run in one shot: audit, then fix, then log. Save the Cat doesn’t fit that model because the decision about what to do is genuinely aesthetic and depends on context I can’t fully specify in advance. The pass can identify whether a moment is needed and generate plausible options. It can’t know which option fits the chapter’s tone or whether I want to go a different direction entirely.

The proposal phase forces a conversation that produces better outcomes than the pass deciding on its own. It’s the same reason we use proposition documents before making cross-chapter architectural changes – big aesthetic decisions should involve the author, not just the AI.

- Zach