---
title: 'In the News: October 26, 2026 (Extra 2)'
description: 'INIT Labs published an open-source coding agent that rewrites its own harness, gated by tests and a rollback layer the model cannot touch.'
canonical_url: 'https://darkfactory.dev/news/2026-10-26-extra-2'
markdown_url: 'https://darkfactory.dev/news/2026-10-26-extra-2.md'
collection: news
date_published: '2026-10-26T16:05:00-04:00'
date_modified: '2026-10-26T16:05:00-04:00'
---

# In the News: October 26, 2026 (Extra 2)


Two items this afternoon: a coding agent that edits its own prompts, tools and compiled core under a rollback guard, and a solo builder's account of where design belongs in an issue-to-pull-request pipeline.

## 1. An open-source coding agent rewrites its own harness and keeps a signed rollback record

**[INIT, the coding agent that rewrites its own harness](https://initlabs.si/)** · INIT Labs · Project page, current as of October 26, 2026

INIT Labs describes the harness as the prompts, tools and loop around a model, and says INIT rewrites its own, tests each change and swaps it in mid-conversation. A component it calls the Root holds the API keys, the signed record and the undo, and the model cannot write to it. The page says INIT v0.3.3 has 136 files, of which 7 are prompts and skills, 5 are tools and 124 are compiled core, and that three changes have been made to them: two kept (October 20 and October 24) and one rolled back (October 21).

The vendor reports these figures, with the logs linked from its page: a core swap in 4.3 seconds, measured end to end with a simulated model; 87.9% of the prompt cache kept across a swap, from a live run on DeepSeek; and a first core change in 13.1 minutes with all checks included. A candidate change must pass four gates, with build and tests taking 439 seconds in the example shown. For the one rollback on record, the page says a change to the shell tool scored 5 points lower, inside the noise, and the gate refused it. The change was rolled back 9 seconds later, and the signed record shows sequence numbers 88 to 90. The page itself notes this was "refused, not proven harmful."

Only the project page has been read. The linked logs were not opened, and the page labels its 21-day comparison with a fixed harness "illustrative, not measured."

**Why it matters:** Most harness work today is a person editing prompts and tool definitions by hand. INIT gives a concrete design for letting the agent do that edit itself: a test gate, a twelve-signal list for rollback, and a record the model cannot alter. Whether the changes it keeps are improvements is the open question. The one refusal on record rested on a 5-point score difference the page itself calls inside the noise.

## 2. A solo builder says the issue-to-PR factory skips design, and that agent sessions still lack memory of past decisions

**[The software factory forgot about design](https://atifaz.am/writing/the-software-factory-forgot-about-design)** · Atif Azam, independent product builder, formerly Coinbase · atifaz.am, October 25, 2026

Azam says he has shipped almost a dozen products in two years and works through four stations: a Linear issue that must state why the work is being done and what done looks like, a design step in Paper where the agent drafts through MCP and he reviews, a coding agent (mostly Codex, he says) that builds against both, and his own review of the result against his intent. His claim is that most software-factory descriptions leave design out, and that "code is cheap now" while deciding what to build is not.

The part most useful to a harness builder is the gap he says remains open. Each agent session gets the issue and the design, but not why an earlier flow was pulled, what analytics showed after a page changed, or why earlier screens were discarded. He calls for persistent memory for product decisions and says nobody has solved it well.

This is one practitioner describing his own workflow. The post gives no measurements of defects, time or cost.

**Why it matters:** The issue template he describes, why plus what done looks like, is a cheap spec gate you can adopt today. His missing-memory complaint is a builder reporting, from product work rather than a benchmark, that new agent sessions start without the history behind earlier decisions.