Welcome to the 317th edition of Coding Jag brought to you by TestMu AI!👐
The creator of Linux has posted a set of kernel patches and said plainly how they were made. The code, the test programs, and the first draft of his write-up were done with Claude Code, under his direction. The gains were measured on a single workload on his own laptop, and as of 13 October, the series had not been merged.
The rest of this edition covers: once AI writes more of the code, who checks it? Anthropic now merges over 30,000 changes a day. A study of more than 700 firms found that pull requests increased by 23 percent after adopting AI coding agents, with no statistically significant change in issue and epic resolution rates. One team bars its coding agents from weakening a test to make it pass. For automation suites, Selenium 4.51 and Cypress 16.2 are out, and Appium 4 is in beta. Also: Python 3.15 makes UTF-8 the default, GitHub Actions is previewing lockfiles for workflow dependencies, and the Deno team is joining Cloudflare. Closer to home, Gartner named TestMu AI a Leader in the 2026 Gartner® Magic Quadrant™ for Agentic Software Quality Assurance Platforms.
📬 Come across something useful or interesting? Just reply and let's exchange ideas.
News
04 min
zdnet.com
🐧 Steven Vaughan-Nichols reports on an eight-patch kernel series whose code, test programs, and first write-up draft Linus Torvalds says were done with Claude Code under his direction. The gains come from a small read workload on his laptop. Torvalds calls memory reclamation the least-solved part, and as of 13 October, the series had not been merged into mainline.
05 minthehackernews.com
🔓 Swati Khandelwal reports that XBOW researchers directed AI agents running OpenAI's public GPT-5.6 Sol model to build a three-bug Chrome attack chain in six days, after months of earlier work. Google fixed all three bugs and paid its top reward of $250,000. The fix only applies after Chrome restarts, so relaunch it.
09 minsocket.dev
🕵️ Socket's research team traced security-audit.yml, a workflow two compromised maintainer accounts committed into 346 repositories on 8 October. A later update counts more than 500 accounts and tens of thousands of repositories. The workflow steals GitHub Actions secrets and also sweeps full git history for cloud and AI keys, so rotating Actions secrets alone is not enough.
03 mindeno.com
🦕 Ryan Dahl announces that the entire Deno team is joining Cloudflare. The runtime gets another year of monthly bug-fix and security releases, after which the team ends its own development, though Deno remains open source. Deno Deploy runs for six more months before shutting down. If your tests or scripts run on Deno, plan ahead.
05 mintestmuai.com
🏆 Gartner has named TestMu AI a Leader in the 2026 Gartner® Magic Quadrant™ for Agentic Software Quality Assurance Platforms, published 12 October. As AI agents move into production, the question shifts from whether a test passed to whether you can trust what an agent says to customers and does on your behalf.
AI
09 minaddyo.substack.com
✅ Addy Osmani argues that with today's models, the bottleneck is no longer writing code but verifying it. Given only one kind of test, he would pick end-to-end tests because a mock-based unit test can pass even when the agent and the test share a misunderstanding. He also warns that flaky suites teach agents to retry.
12 minanthropic.com
🔀 Anthropic went from around 2,500 merged changes a day in May to over 30,000 by September. A seven-person team built Scion, an in-house version control system, and moved the monorepo off git in five months. Unable to replicate a global merge-queue lock at agent scale, Anthropic invested in automated testing for post-merge and pre-deploy stages.
05 minarstechnica.com
📦 Kyle Orland covers a study of more than 700 software firms. After AI coding agents arrived, lines of code rose 30 percent and pull requests 23 percent, but issue and epic resolution rates showed no statistically significant change. Time from pull request to merge increased by 49 percent, and the share of pull requests requiring changes nearly doubled.
09 mintestmuai.com
🧾 In an Anthropic test, a model told to stop before the final step mistakenly submitted the form several times. It expected another confirmation page, and its reasoning said it was only demonstrating the process. Anthropic says the cases so far had minimal real-world impact. For testers: grade requests sent, forms submitted, and records changed, not the agent's account.
15 mincockroachlabs.com
🏥 Adam Storm and Rafi Shamim share the rules behind Cockroach Labs' agent pipeline. An agent may never disable, skip, or weaken a test to make it pass, because a failing test is correct until proven otherwise. Every review starts by disabling the change and checking that the new tests fail. Five months brought only seven reverts.
Automation
03 minselenium.dev
🧪 Diego Molina announces Selenium 4.51 for JavaScript, Ruby, Python, .NET, Java and the Grid. In Java, Guava is no longer a dependency of the client bindings, so check the migration guide if your code relies on ExpectedCondition as a Guava Function. Python 3.15 is supported, and the JavaScript driver can install extensions through BiDi.
08 mincypress.io
♿ Jennifer Shehane introduces cy.getByRole(), a built-in Cypress 16.2.0 query that finds elements by role and accessible name. It matches native elements by default, so if someone swaps a real button for a div with a button role, the test fails and points at the change. Teams can migrate from Cypress Testing Library one spec at a time.
12 minthegreenreport.blog
💳 The Green Report builds a Playwright suite for a webhook receiver, starting from the rule that delivery is at least once, so duplicates are normal. A receiver that passes a happy-path test can still charge a customer twice. Seven deliberate mutations each turned at least one test red, and one was caught only by the simultaneous-delivery test.
12 mintestmuai.com
📱 Appium 4 is in public beta with no release date yet, and betas land weekly. It is a modernization release, not a feature release: Node 22.22 or newer, ESM-only packages, and the last of the old JSONWP protocol is gone. For most suites, upgrading is a Node bump and a reinstall. The post lists what breaks.
Tools
03 minblog.python.org
🐍 Hugo van Kemenade announces Python 3.15, built from 5,643 commits by 1,012 contributors. UTF-8 is now the default encoding, explicit lazy imports speed up startup, and the experimental JIT beats the standard interpreter by 7 to 8 percent on x86-64 Linux. A new default encoding is a good reason to run your tests before upgrading CI images.
02 mingithub.blog
🔒 GitHub Actions can now lock a workflow's dependencies, in public preview on Enterprise Cloud, Team and Free plans. Tags and branches can move after you review a workflow, so a later run might execute different code. A lockfile records reviewable commits for actions, their transitive dependencies, and reusable workflows. Locking does not make a dependency safe.
15 minbun.com
🥟 Jarred Sumner introduces bun check, a TypeScript type checker built into Bun that passes all of TypeScript 7.0.2's conformance tests. Run bun test --check, and a type error in a test file or in anything it imports stops the run before any tests start. The release also fixes a matcher bug that let a wrong assertion pass.
Video & Podcast
07 minmaintainable.fm
🎧 On Maintainable, Robby Russell talks with Gabriela Moreira, CEO of Quint. She describes a new codebase that grew from roughly 200 tests to 800 with help from AI, while she trusted the suite less. They cover model-based testing, trace validation, and why regression tests that preserve the history of real problems are worth keeping.
12 minyoutube.com
📲 This 44-minute TestMu AI tutorial connects Appium MCP to Claude Code. An AI coding agent can then start Android and iOS sessions, install an app, and drive it with natural-language prompts. It runs a full login-and-swipe flow from a single instruction, then turns that AI-driven flow into a WebdriverIO test script.
Events
05 mintestcon.lt
🎬 TestCon Europe's 11th edition runs in Vilnius, with workshops on 20 October. Three conference days follow from 21 to 23 October at Forum Cinemas Vingis, a cinema venue, with live streaming for online ticket holders. Topics include AI, accessibility, scalable automation and testing at scale. TestMu AI is a sponsor.