---
title: 'In the News: September 3, 2026'
description: 'Developer Scott Fryxell reports cutting Fable usage 75% after adopting a five-role harness and using a cheaper model by default.'
canonical_url: 'https://darkfactory.dev/news/2026-09-03-extra-2'
markdown_url: 'https://darkfactory.dev/news/2026-09-03-extra-2.md'
collection: news
date_published: '2026-09-03T16:15:00-04:00'
date_modified: '2026-09-03T16:15:00-04:00'
---

# In the News: September 3, 2026


Scott Fryxell says a five-role coding-agent pipeline and a cheaper default model cut his use of Fable by 75%. Ryan Lopopolo separately describes harness engineering as a way to provide just-in-time opportunities for in-context learning.

## 1. A developer cut frontier-model usage 75% by giving the harness five roles

**[The Harness Is the Thing](https://scott-fryxell.github.io/blog/the-harness-is-the-thing/)** · Scott Fryxell, independent developer · Aug 25, 2026

Fryxell runs three coding-agent TUIs, Cursor, Claude, and Pi, sharing one set of skills and one `AGENTS.md`, and defaults to the cheaper `deepseek-v4-flash-0731` for maintenance work, reaching for frontier models only when exploring a large feature or refactor. He built that split into a five-stage pipeline: explore, plan into an explicit task DAG, worker, critic, and a fifth stage he added himself, promoter, a forced step for writing up and communicating finished work, which he says he needed because he tends to ship and move on. He also adopted Can Bölük's "prewalk" technique, using a frontier model for planning and the first task, then handing off to the cheap model once the pattern is set. This, he writes, "has dropped my usage of Fable in even my most intense contexts by 75%," a change he says "feels sane, something akin to AI soak." He separately describes routing his own editor configuration through the same harness as "harness as jig."

**Why it matters:** The role split gives readers a specific setup to test: use a frontier model for planning and critique, a cheaper model for execution, and require a write-up at the end. The 75% figure comes from one person's own use and has not been independently measured, so it is evidence from one setup rather than a benchmark.

---

## Also this cycle

- **[A tighter definition of what harness engineering is doing](https://x.com/_lopopolo/status/2095195800510902711)** · Ryan Lopopolo, Principal Engineer, Agentic Google Cloud, formerly OpenAI · Sep 2, 2026. Lopopolo argues the practice is mechanical rather than motivational: "All of 'harness engineering' is essentially tricks to provide JIT opportunities for ICL to align model behavior with what good looks like for you..." The post had 3,136 views and 52 likes at 26 hours, modest reach for this framing.