Welcome to the 313th edition of Coding Jag brought to you by TestMu AI!👐
Your AI doesn’t know what correct code is. Picture a team asking an assistant to build a buy one, get one 50% off promotion. The code works, and it also lets shoppers stack discounts and collect loyalty points nobody authorized. None of that was requested. The rules forbidding it live in a test suite the model never read.
The same gap ran through the week, between code that lands in minutes and checks that still take judgment. One engineering lead now calls the review queue the new backlog. An agent cut off from the web still passed its evaluation by quietly reading a source checkout on the same machine. Elsewhere, Java 27 landed, and Shopify rebuilt its flagship app in Swift and Kotlin.
Our new TestMu AI Build vs Buy: AI Testing Agent Whitepaper is out. It prices both paths over year one: what an in-house agent built on coding agents costs, what a purpose-built platform costs at list price, and how to make the call. Written for engineering and QA leaders. Free to download.
📬 Come across something useful or interesting? Just reply and let’s exchange ideas.
News
07 min
infoq.com
📱 Bruno Couriol reports that one engineer’s proof of concept took a week, and then six engineers rebuilt the flagship Shop app from the ground up, deploying to production in 12 weeks. Cold start improved by 23% on iOS and 50% on Android. Session stability went from 99.5% to 99.95%.
06 mininside.java
☕ The Java team ships nine enhancements big enough to warrant their own proposals. Four are final: post-quantum hybrid key exchange for TLS 1.3, G1 as the default garbage collector everywhere, compact object headers by default, and in-process redaction for flight recorder data. Oracle updates this one until March 2027, when JDK 28 takes over.
08 minthehackernews.com
🕵️ Ravie Lakshmanan covers Anthropic’s 154-page threat report. One group ran credential harvesting across 10 cloud workers, scanning 1.8 million Android app packages. Another hit roughly 50 organizations. The report’s own summary is the part worth sitting with: at the far end, operations ran autonomously, with minimal human input or supervision.
05 minsimonwillison.net
💎 Simon Willison walks through a four-author report that makes the case for OpenAI agents running the May attack on Ruby’s package registry. An "oai" marker turned up in package names, author fields, and a fake email address, and the code inside appeared to be model-written.
10 minsurveys.lambdatest.com
📊 Our annual survey on how teams actually use AI in testing and quality engineering is still open. Roughly ten minutes to complete. Every response is handled with the utmost confidentiality, and we publish the results, sharing the most impactful insights with the global quality community. Honest answers sharpen the picture for everyone.
AI
11 mingithub.blog
🦀 Stephen Toub, a Distinguished Engineer at Microsoft, documents 832,378 lines of production Rust replacing about 430,000 lines of TypeScript, across 128 pull requests in roughly 14 weeks. Every existing end-to-end test ran against the new code at each step. Three families of regression dominated, including tests that asserted the wrong thing.
07 mintechnode.global
📥 Kevin Lee, Head of Deployed Engineering APJ at Cognition, says that inside his own organization, more agent sessions now start from events and schedules than from a person. His line: the review queue becomes the new backlog. His fix is three habits, starting with a test plan written against the real source before testing begins.
08 mindeveloper.microsoft.com
🧪 Waldek Mastykarz, Principal Developer Advocate at Microsoft, blocked web access to test what a model knew, and the answers still came back right. The agent had found a local install and its source checkout, then read the version tag. A valid answer can still produce an invalid measurement. Check the trajectory, not the score.
10 mindatadoghq.com
🚦 Candace Shamieh, T Zhang and Gabriele Baldoni at Datadog land the sentence platform teams need: a well-behaved agent doesn’t guarantee a well-behaved platform. One internal coding agent workload was estimated at 2,000 requests a minute and 45 billion tokens a day. They stretched gateway shutdown windows to 60 seconds so in-flight requests survive rolling deployments.
07 minthoughtworks.com
📈 Nik Malykhin served as the AI champion for eight months for a team of six developers at Thoughtworks. The baseline was about 15 user stories per iteration. It fell to 12 as everyone adjusted, then rose to 27 once they settled on a simpler workflow. The dip came before the gain, which is the useful part.
Automation
08 minsoftwaretestingmagazine.com
🎯 Hélder Ferreira, Director of Product Management at Sembi, works through an example: a buy-one-get-one-50% off feature that would let shoppers stack discounts and collect loyalty points without authorization. The model never saw the rules forbidding it. His fix: feed tests to the assistant as context, because that is where real behavior is written down.
06 mingithub.com
⚙️ The keyword-driven automation framework ships a feature release, and Libdoc gets the attention. Library documentation can now be written in Markdown, and arguments, return values, and exceptions can be documented in Google style, rendered alongside automatically collected names and types. Tests and tasks also accept their own metadata, and console logging is configurable.
10 mindev.to
🧹 Mike Ralduxin of DeviQA audits a broken suite across four layers: test design and data, framework currency, execution infrastructure, and ownership. His rule is blunt: one failing layer means refactor, two or more means rebuild. Cited research puts async waits at 45% of flaky causes, with 75% of flaky tests clustering around a shared root cause.
Tools
05 minnodejs.org
⚙️ Antoine du Hamel cut this one, and the headline for testers is node:bench, a benchmarking module now in the standard library, though still behind an experimental flag. Web Workers arrive in the worker module, the foreign function interface is on by default, and the virtual file system plugs into both module loaders.
06 mingithub.com
🧹 Eight new lint rules land, and two of the changes target test code. One rule enforces valid titles for unit tests and suites, and separately, the analyzer now recognizes suite() as an alias for describe(). Another rule stops you from returning from inside a promise finally callback.
06 minblog.postman.com
⏰ Harsh Vardhan walks through monitors that now schedule gRPC and GraphQL calls, from every five minutes up to weekly. Existing auth, variables, and test scripts carry over without edits. Two gaps to know before you plan around it: GraphQL subscriptions are unsupported, and client streaming takes one message per call.
Video & Podcast
07 minse-radio.net
🎙️ Host Giovanni Asproni talks to Milan Milanovic, a Microsoft MVP for Developer Technologies, about 53 empirical and mathematical laws of software engineering. Conway’s and Brooks’s are there, alongside less-quoted ones from Gilb and Tesler. The thread worth following is what those laws mean for teams now leaning on AI tools.
12 minyoutube.com
🧠 The AWS Developers channel spends 22 minutes on what belongs in an agent’s context window, and what should be thrown out. Four patterns: compress what is old, externalize what is large, select what is relevant, isolate what does not belong. Working code throughout, plus a one-liner for sensible defaults.
Events
05 minwrobocon.eu
🤖 A free online conference on 8 October, marking its fifth year, built entirely around Robot Framework automation. Speakers include Tim de Groot of TestCoders, Asko Soukka of the University of Jyväskylä, and Ahmed Hassan of NTT DATA. Wrobocon 2026’s theme is "Robot Framework University," and it is organized by the Polish community and open globally.
06 mintokyotestfest.com
🗼 The conference runs on 13 November at Osaki Bright Core Hall in Tokyo, with two parallel tutorial tracks on 12 November at the same venue. Angie Jones opens, and Richard Bradshaw closes. It is designed so that Japanese- and English-speaking testers can meet international speakers in one room.