← Newsletters

React and Try Something Different (rwblog S6E13)

Last updated: Wed Nov 15 2023


A REPL for Writing

A possibly-productive metaphor between software engineering and writing: is there a benefit to a faster “REPL loop” in fiction writing?

To explain for the non-programmers in the audience (are there any?): some programming languages come with a “read-eval-print loop”, or REPL, which lets you write a line of code at a time and immediately see what it does. If you’ve ever run python3 on the command line or typed JavaScript in the browser console, that’s a REPL!

The benefit is that you get immediate feedback about what does and doesn’t work while programming, which is great, because speed matters.1 If something doesn’t work, you can immediately react and try something different. This is also why programmers tend to care so much about fast compile times (even if slow compile times are a great excuse for slacking).

(Digression: This is why ChatGPT’s Code Interpreter is so exciting, as Simon Willison has shown. It basically puts ChatGPT and a Python interpreter into one giant REPL loop that you can interact with.)

So, the metaphor: I recently finished a 65k-word novel draft, and, per Matt Bell’s Refuse to Be Done (highly, highly recommended for anybody planning to write a novel), I’m now about to throw the whole thing out and rewrite it from scratch. Now, that’s a great exercise — I already see so many things I want to change — but it does make me wonder if there’s an opportunity to deliberately practice improving the feedback time.

Now, a novel is maybe not the best example, since so many of those 65k words were generative, “writing is rewriting,” etc. But on a smaller scale, were all of those 65k words actually useful? Similarly, in the past I’ve certainly dropped a few thousands words into a short story only to realize it was just fundamentally flawed. Was there a way I could have noticed that earlier and pivoted to a different idea?

Some ideas:


  1. Although Hillel Wayne has a contrary take — thanks to Goodhart’s law (“when a measure becomes a target, it ceases to be a measure”), slow feedback loops can sometimes be a very good thing!