---
title: 'In the News: September 25, 2026 (Midday)'
description: 'Geoffrey Huntley stakes a third definition of "software factory" on X, and a new Claude Code memory tool claims sub-second, cent-level decisions.'
canonical_url: 'https://darkfactory.dev/news/2026-09-25-midday'
markdown_url: 'https://darkfactory.dev/news/2026-09-25-midday.md'
collection: news
date_published: '2026-09-25T13:15:00-04:00'
date_modified: '2026-09-25T13:15:00-04:00'
---

# In the News: September 25, 2026 (Midday)


Two items today. A fresh voice enters the definitional fight over what a "software factory" actually is, and a new open-source tool tries to make agent memory a cheap, automatic per-turn decision instead of an occasional summarization job.

## 1. Geoffrey Huntley: "your product is the factory"

**[Geoffrey Huntley on X](https://x.com/GeoffreyHuntley/status/2103470149450022989)** · Geoffrey Huntley, independent practitioner, creator of the "Ralph" agent loop pattern · X, September 25, 2026

Posting his "hottest take" on what a software factory actually is, Geoffrey Huntley wrote: "A software factory is actually an applied automation practice in the actual product. It's not some external thing. It's a product pattern where the product builds the product inception style. Any external system or dependency should be internalized into the product to enable the product to build the product. If you can't build the product in the product, from the product, then you are missing the mark." He closed with a one-line summary: "Your product is the factory." A same-day follow-up reply added: "the factory is the widget, it's not a something that produces a widget which can be used to build something." The post had drawn 129 likes, 26 reposts and 7,553 views about four hours after posting. Huntley captioned an attached image "a sneak preview behind an embedded software factory," pointing to a fuller post on his own site that has not published yet.

Huntley's post is the third distinct definition of "software factory" to surface from a named voice in about a day, following Matt Pocock's AI Coding Dictionary entry (covered in this morning's edition) and an in-thread amendment from Dex Horthy arguing the term should also cover human-shepherded changes with automated checks attached. Pocock's definition includes the surrounding review and deployment automation around a product; Huntley's is narrower and stricter: the automation has to live inside the product itself, not sit beside it as external tooling.

**Why it matters:** Three named practitioners publicly disagreeing about what a "software factory" is, inside about a day, is worth tracking rather than settling prematurely. Huntley's specific claim, that a factory built as an external system misses the point, is a real fork in the road for anyone deciding whether to build agent tooling into the product itself or as a separate harness layer.

## 2. A memory tool for Claude Code claims sub-second, cent-level decisions

**[jevmem](https://github.com/Avinash-jetwani/jevmem)** · Avinash Jetwani · GitHub, accessed September 25, 2026

jevmem is a new open-source tool that automatically saves decisions, constraints, bugs and todos from Claude Code sessions into a JEVMEM.md file, then re-injects the relevant lines into the next session's context. Rather than asking a large model to judge what is worth remembering, jevmem routes each turn through a small classifier service, called Jev, that scores it in about 300 milliseconds, and applies fixed thresholds written in code rather than a prompt. In the author's own benchmark of 66 held-out turns, jevmem matched GPT-6 Astra's 98.5 percent save-or-skip accuracy at roughly 300 milliseconds and $0.000127 per decision, against 2.8 to 4.3 seconds and $0.005 to $0.013 per decision for six current models asked to make the same call directly. On the finer save-plus-kind judgment, jevmem trailed GPT-6 Astra and Claude Opus 5.5 by a few points.

The author states plainly that this is not an independent result: "both eval sets were written by the author, and neither is an independent benchmark," and the tool is labeled v0.4 and "early." Using it means sending each turn's text, the prior two turns, and existing memory lines to Jev's API for scoring, after stripping common secret and email patterns first; a saved memory line is then optionally written using an OpenAI or Anthropic key supplied by the user.

**Why it matters:** The bet here, that a fast, cheap, threshold-based classifier can do nearly as well as routing every turn through a frontier model, is a concrete answer to a cost and latency problem teams running agents at scale actually have. It comes with a tradeoff its own author states clearly: chat text leaves the machine to be scored by a third party, and the benchmark backing the claim is self-graded rather than independently verified.