---
title: 'In the News: November 11, 2026 (Morning)'
description: "Anthropic's Fable 5.2 prompting guide measures how one sentence about code ownership changes whether a coding agent merges duplicated code."
canonical_url: 'https://darkfactory.dev/news/2026-11-11-morning'
markdown_url: 'https://darkfactory.dev/news/2026-11-11-morning.md'
collection: news
date_published: '2026-11-11T07:10:00-05:00'
date_modified: '2026-11-11T07:10:00-05:00'
---

# In the News: November 11, 2026 (Morning)


Anthropic's prompting guide for Claude Fable 5.2 reports what specific system prompt lines do to an agent, with rates from Anthropic's own testing. In Anthropic's tests on small coding tasks, telling the model the codebase is its own moved repeated-code merges from under a tenth of tries to about half, at a cost in tokens and unrequested changes. The second item is a Thoughtworks build of one procurement agent on two durable execution runtimes, which handled the same mid-task rule change in opposite ways.

## 1. Anthropic measures what a code-ownership sentence does to Fable 5.2

**[Prompting Claude Fable 5.2](https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/prompting-claude-fable-5-2)** · Anthropic · Claude Platform documentation, quoted by [Simon Willison](https://simonwillison.net/2026/Nov/10/prompting-claude-fable-52/) on November 10, 2026

According to the guide, when a prompt said nothing about ownership or presented the code as someone else's, the model "tended to add to the existing structure, even when that repeated existing code." Anthropic's suggested line is "This codebase is yours: what it's for and how it's built are your call, and you'll keep building on it." On small coding tasks, the model merged repeated code in about half of the tries with that sentence and in under a tenth without it. The sentence has costs. Output tokens rose about 1.5 times, and the model made changes nobody asked for, many of them new tests and documentation. Apart from bug fixes, it changed how an existing feature behaves in one try in six to ten, more often at higher effort. A "No drive-by refactors, renames or cleanups" rule in the same prompt still left merges in about a third of tries.

The guide measures other harness text the same way. An approval rule that listed "sending messages to people outside this conversation" made the model hold back roughly half of its routine requests to colleagues in simulated tasks with nobody there to approve. One stale "never retry" note, among a dozen notes from earlier sessions, cut retries after a busy tool to about four in ten, from nearly all. Text telling the model to finish the whole task cut sessions that ended with doable work left to about half, in replays at medium effort, and raised output tokens by about 30 percent. All of these are Anthropic's figures from its own tests.

**Why it matters:** The guide treats prompt lines and accumulated agent notes as switches with measured side effects, so a harness carried over from an earlier model needs retesting, not just a new model name. If you add the ownership sentence to get cleanup, review the diff, because Anthropic says the model's closing summary "often left out small ones."

## 2. Temporal and Cloudflare split on a rule change under a waiting agent

**[Long-horizon agents and durable execution](https://martinfowler.com/articles/durable-execution-agents.html)** · Kiran Prakash, Director of Data and AI, Thoughtworks · martinfowler.com, November 10, 2026

Prakash built Stock Replenisher, an agent that gets supplier quotes, waits days for a manager to approve a price, and follows the order to delivery, on both Temporal and Cloudflare Workflows. Durable execution records the result of each external step, so after a crash the loop replays without asking for new quotes or a second approval. A section titled "Durable is not the same as correct" covers what replay does not fix. A lost order acknowledgement looks like a failure, so the agent orders again from another supplier; his answer is a cancel_order tool. Approvals go stale while the agent waits, so the manager approves a price up to 5% above the quote and the place_order tool re-reads the quote before it commits.

His clearest result came from a deploy. While a task that had drafted a purchase order from three quotes waited for approval, he shipped a rule requiring four. "On Temporal, SKU-123 silently continued with the order as per the old rule, which the new rule doesn't permit." On Cloudflare, where his loop re-checks recorded decisions against current rules, the order stayed a draft. His fix records a rules version in the task's state and branches on it. Both behaviors are reported from the author's own build.

**Why it matters:** In a harness that suspends tasks for days, every deploy is a migration against in-flight work, and the runtime decides whether the old rule wins quietly or the task stops. Test which one yours does before shipping a rule change. Prakash also keeps the business rules in tool code rather than the prompt, so a supplier email claiming approval "approves nothing."