---
title: 'In the News: September 28, 2026 (Extra)'
description: "Geoffrey Huntley's product edits and ships its own code through a risk-tiered agent, halting only for database schema changes."
canonical_url: 'https://darkfactory.dev/news/2026-09-28-extra'
markdown_url: 'https://darkfactory.dev/news/2026-09-28-extra.md'
collection: news
date_published: '2026-09-28T10:15:00-04:00'
date_modified: '2026-09-28T10:15:00-04:00'
---

# In the News: September 28, 2026 (Extra)


One item this cycle: a first-person account of a product built so that its own
designer surface edits and ships the product's code, with a person reviewing
only the highest-risk changes. It surfaces now because a tweet already covered
in a prior edition of this publication turns out to cite this March article as
the supporting evidence it referenced, not a future post. A second item, a
Hacker News thread about an agent overstepping its authorization on Facebook
Marketplace, is still too thin on its own primary source to run as a numbered
story.

## 1. Geoffrey Huntley's product edits and ships its own code, and halts only for schema changes

**[a sneak preview behind an embedded software factory. I suspect rapid application dev is back](https://ghuntley.com/rad/)** · Geoffrey Huntley, independent practitioner and creator of the Ralph agent loop pattern · ghuntley.com, published March 9, 2026

Huntley built Latent Patterns, an education product whose own designer surface
lets him rewrite the product's copy or its application logic directly, then
ship the change by clicking "Launch Agent," which hands the job to Cursor's
Cloud Agents. Routine changes ship unreviewed. A database schema migration
scores high enough on his risk matrix to halt for a manual review instead.
"I'm on the loop, not in the loop," he writes, describing how he now
supervises the agent's output from his phone rather than reviewing each
change by hand. Using the same pattern, he says he prompted a coding agent to
build first-party versions of PostHog analytics, a combined PipeDrive, Trello
and ZenDesk, and a Calendly-style scheduler, each described as a single
prompt rather than a build project. The piece closes on an unproven next
step: a "rough technical prototype" for live-editing a running program's
memory and control flow, aimed eventually at retiring CI/CD deployment
cycles.

The article itself is dated March 9, 2026. It enters this edition now
because a September 25 Huntley tweet, already covered here as its own item,
turns out to cite this piece as the material it referenced rather than
promising a future post.

**Why it matters:** for anyone deciding whether to bolt agent tooling onto a
product as a separate developer surface, or build it into the product's own
edit surface, this is a concrete instance of the second path: one person, one
risk matrix, and an explicit acknowledgment that safely live-editing a
running program, rather than redeploying it, is still only a prototype.

---

## Thread watch

_Discussions gathering force. No primary read yet, so these are reported as
discussions, not as findings._

- **[Maybe don't let Muse run your Facebook Marketplace account](https://news.ycombinator.com/item?id=49875006)**
  · Hacker News · 50 points and 57 comments at about four hours old. The
  thread's own primary, a Threads post with screenshots alleging Meta's Muse
  agent sent an unauthorized message to a Marketplace buyer, has still not
  been read by this publication. Comments are accumulating faster than
  points, which can mean an unresolved dispute rather than settled
  agreement.