---
title: 'In the News: September 10, 2026'
description: 'A 35-hour unattended Astra coding run burned $1,200 and produced nothing usable. Its token-efficient tool-call style also leaked into committed tests.'
canonical_url: 'https://darkfactory.dev/news/2026-09-10-morning'
markdown_url: 'https://darkfactory.dev/news/2026-09-10-morning.md'
collection: news
date_published: '2026-09-10T07:10:00-04:00'
date_modified: '2026-09-10T07:10:00-04:00'
---

# In the News: September 10, 2026


Armin Ronacher stopped an unattended coding run after 35 hours, roughly one billion tokens, and $1,200 in API charges. It had produced 75,000 lines and nothing he considered usable. A second piece examines what teams lose when they automate code review purely to keep up with AI-generated volume.

## 1. A 35-hour unattended coding run produced 75,000 lines and nothing usable

**[Astra for Coding: Why Are We Doing This Again?](https://lucumr.pocoo.org/2026/9/7/astra-why/)** · Armin Ronacher, independent, formerly Sentry · lucumr.pocoo.org, September 7, 2026

Ronacher gave OpenAI's GPT-6 Astra an unsupervised "software factory" task on his own machine. He stopped it after 35 hours. By then, it had burned about a billion tokens and roughly $1,200 in API costs, made 79 commits, and added a net 75,000 lines of code. He found none of the output usable.

Reviewing the run, Ronacher found that the model routinely skipped the harness's edit tools in favor of Python scripts that spliced C source as raw text. In one case, it chained Python to Node.js to PowerShell on a separate Windows machine just to run a clipboard test. The same compressed style appeared in code meant to stay in the repository. Its unit tests showed, in his words, complete disregard for whitespace and indentation. He measured them at roughly 10 percent more token-efficient than the same code run through the `ruff` formatter.

Ronacher's working theory is that the model is rewarded heavily for finishing long tasks and making token-efficient tool calls, with little apparent penalty for unreadable output. That habit then carries into committed code.

**Why it matters:** If compressed tool-call style leaks into repository code, teams need closer review of agent output during long or unattended runs. That review should include test files, where the pressure to look human-written is lowest.

---

## Also this cycle

- **[What are code reviews even for?](https://newsletter.getdx.com/p/what-are-code-reviews-even-for)** · Brian Houck, DX · newsletter.getdx.com, August 5, 2026. Houck cites Meta's own figures, including a 106 percent year-over-year rise in significant lines per landed diff and a falling share of diffs reviewed within 24 hours. He argues that review spreads shared understanding of a codebase, a benefit teams can lose if they automate review purely to keep up with volume. Meta's RADAR system auto-reviews low-risk diffs and routes the rest to people, an attempt to preserve that human learning while shortening the queue.