---
schema: 1
type: term
title: '**Claude 3.5 Sonnet**'
description: >-
  The LLM credited by Val Town founder Steve Krouse as the most important thing
  behind Townie's ability to rebuild a working full-stack app from a single
  prompt in under a minute.
canonical: >-
  https://liquidgold.cc/channel/everyinc/explore/term/claude-3-5-sonnet-718991f7f1d3
markdown: >-
  https://liquidgold.cc/channel/everyinc/explore/term/claude-3-5-sonnet-718991f7f1d3.md
llms: null
parent: 'https://liquidgold.cc/channel/everyinc/llms.txt'
channel: everyinc
updated: '2026-10-23T22:12:35.346Z'
language: null
children: []
---


# \*\*Claude 3.5 Sonnet\*\*

Canonical: https://liquidgold.cc/channel/everyinc/explore/term/claude-3-5-sonnet-718991f7f1d3

Parent: [Channel index](https://liquidgold.cc/channel/everyinc/llms.txt)

## Definition

The LLM credited by Val Town founder Steve Krouse as the most important thing behind Townie's ability to rebuild a working full-stack app from a single prompt in under a minute.

In this channel's material, **Claude 3.5 Sonnet** appears as the underlying model behind Val Town's AI assistant Townie — and as the centerpiece of founder Steve Krouse's explanation of why AI coding jumped forward in a single year. [^ebfee90b-4a43-55ab-9be2-5070c4aaefb0@0] Krouse credits three things for that leap: integration (the copy-paste loop between an LLM and a coding platform becoming one product), the system prompt (telling the model code runs in Deno JavaScript, that no server needs starting or deploying, and which platform bugs to avoid), and — in his words — **the model itself**: Claude 3.5 Sonnet was the most important thing that happened. [^ebfee90b-4a43-55ab-9be2-5070c4aaefb0@0] The episode also uses the model to explore two broader industry themes: - **Perceived degradation and evals.** Before Val Town had real evals, users reported the model getting _stupider_ over weekends, and Krouse admits helping fuel a brief industry panic where everyone convinced each other Claude 3.5 Sonnet had degraded. Once evals existed, he could check complaints against them — and confidently say _try again_ if nothing showed up. - **Staying neutral to model progress.** Val Town architected a model switcher, beta flags, usage-based pricing, and per-model prompting so it could jump when a new Sonnet shipped — the way Cursor and WebSim did. In practice they actually **rolled back** from the newer Sonnet because they couldn't get its prompting right, showing that harness quality matters as much as the model name. [^ebfee90b-4a43-55ab-9be2-5070c4aaefb0@0] Use this term when the material discusses which model powers an AI coding tool, model-driven leaps in capability, or the architecture needed to benefit from a new model without rearchitecting a product.