---
title: 'In the News: October 29, 2026 (Extra 3)'
description: 'Thoughtworks ran a week of fully agent-built software; the app ran but the build stayed red, and the open question is whether CI can absorb parallel agents.'
canonical_url: 'https://darkfactory.dev/news/2026-10-29-extra-3'
markdown_url: 'https://darkfactory.dev/news/2026-10-29-extra-3.md'
collection: news
date_published: '2026-10-29T23:05:00-04:00'
date_modified: '2026-10-29T23:05:00-04:00'
---

# In the News: October 29, 2026 (Extra 3)


Thoughtworks says a week of fully agent-built software produced a running application and a frequently red build. Separately, a YC Labs experiment that put swarms of Codex and Claude Code agents against each other found a misconfigured defense only after the team lost a game.

## 1. Thoughtworks: a week of agent-only development shipped a running app and a build that was often red

**[Why are good developers so wary of agentic development?](https://martinfowler.com/rachels-ramblings/good-devs-wary.html)** · Rachel Laycock, CTO, Thoughtworks · martinfowler.com, October 29, 2026

Laycock describes an internal experiment run by Giles Alexander, Thoughtworks' Regional CTO for Europe. Senior engineers, infrastructure consultants and product people spent a week building an airline application with no code written by hand. She says they got about 80% done. The team's own estimate was about two thirds of their time at the whiteboard and one third on the harness, and the agents kept producing code after the working day ended.

The application ran, "but the build was frequently red." Laycock lists candidate causes: too many agents working in parallel and stepping on each other's changes, a system not modular enough for separate work, or changes arriving faster than the checks could finish. She says the team has not settled which. The experiment ran twice with different teams. She also cites DORA's 2025 research as linking AI adoption to higher delivery throughput and lower delivery stability.

**Why it matters:** This is one firm's account of two runs, with no published data, so treat the time split as an estimate. The useful part is where it broke. A red build only works as a signal if it stays rare, and her own question about batch size, with teams moving from stories to epics, is the one to test before you add more parallel agents.

## 2. YC Labs ran Codex and Claude Code swarms against each other; one side loaded 55 of its 74 anti-air turrets with the wrong ammunition

**[Agents in Games](https://labs.ycombinator.com/projects/games/index.html)** · YC Labs, with Olam Labs · October 2026 (discussed on [Hacker News](https://news.ycombinator.com/item?id=50296486), 190 points about 5.5 hours after posting)

In Mindustry, a factory-building strategy game, GPT-6.1 Sol agents in Codex beat Claude Opus 5.5 agents in Claude Code 6-1 across seven games over about 9.5 hours. In Bed Wars, 10 agents per side, the GPT team won 6-3 in nine games. The authors report that each agent got a generic harness and the game API, with no custom memory work and no instructions on how to play.

The failure most relevant to software teams came in Mindustry game 6. After losing to a 24-zenith air strike, a Claude agent found that "55 of them were loaded with COPPER," meaning 55 of the team's 74 anti-air duo turrets, where graphite was needed. In game 7 the team built 207 duos fed with graphite, and the authors say that stopped the same strike. In a separate 50-agent Minecraft run, they report 34 agent deaths, 22 of them in the first night, and a colony that built a ration ledger and shifted toward food and survival.

**Why it matters:** The copper-for-graphite miss is a verification failure that a team of agents carried through 55 turrets until a loss exposed it. The model comparison is confounded: it pits two vendors' coding harnesses against each other in one game, and the authors' behavioral readings, such as GPT agents acting "more like a hive," are their interpretation of a small number of games. It is a lead on swarm coordination, not a ranking.

---

## Also this cycle

- **[Mistral Large 4 (Le Chonk) is now open weight](https://mistral.ai/news/mistral-large-4-oss/)** · Mistral AI · Mistral says it is releasing the weights of its 1-trillion-parameter model. It reports 41% on Terminal-Bench 4.0 (up 13 points from the preview), 19.7% on Cognition's private FrontierCode 2.0, and a 48% score on the Artificial Analysis Coding Agent Index, effectively tied with GLM-5.3 at 47% fewer tokens per task. These are Mistral's figures and we have not run them; we did not check the license terms.