---
schema: 1
type: term
title: _YOLOing it_
description: >-
  _YOLOing it_, as used in the channel's captured material: the practice of
  deploying changes to a live AI product with no real **evals** in place,
  testing manually and relying on anecdotal reports — the way Val Town ran
  Townie for its first three or four months before building a measurement
  harness [^ebfee90b-4a43-55ab-9be2-5070c4aaefb0@0].
canonical: 'https://liquidgold.cc/channel/everyinc/explore/term/yoloing-it-92005835b4da'
markdown: 'https://liquidgold.cc/channel/everyinc/explore/term/yoloing-it-92005835b4da.md'
llms: null
parent: 'https://liquidgold.cc/channel/everyinc/llms.txt'
channel: everyinc
updated: '2026-10-23T22:12:35.346Z'
language: null
children: []
---


# \_YOLOing it\_

Canonical: https://liquidgold.cc/channel/everyinc/explore/term/yoloing-it-92005835b4da

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

## Definition

_YOLOing it_, as used in the channel's captured material: the practice of deploying changes to a live AI product with no real **evals** in place, testing manually and relying on anecdotal reports — the way Val Town ran Townie for its first three or four months before building a measurement harness [^ebfee90b-4a43-55ab-9be2-5070c4aaefb0@0].

Explore how **_YOLOing it_** shows up in this channel's captured material: it comes from the episode with Val Town founder Steve Krouse, where he describes the early months of running Townie, Val Town's AI assistant, without any real **evals** in place. For the first three or four months, the team would _YOLO_ changes straight to production, test the result manually, and rely on anecdotal user reports — including weekend panics where users were convinced the model had gotten _stupider_, which Krouse would take to Twitter to cross-check with [^ebfee90b-4a43-55ab-9be2-5070c4aaefb0@0]. In this channel's usage, _YOLOing it_ therefore names the practice of pushing changes to a live, stochastic AI product with no measurement harness to catch regressions — and the episode pairs the term with its remedy: once evals existed, Krouse could check customer complaints against them and confidently tell users _try again_ when nothing showed up. The term also carries a caution he draws from the experience — you start only taking seriously what you can measure, and squishy products still miss a lot that hard-enough evals would catch [^ebfee90b-4a43-55ab-9be2-5070c4aaefb0@0].