Ars Technica · Two major AI coding tools wiped out user data after making cascading mistakes
Administered by:
#aicoding
11 posts8 participants1 post today
Two major AI coding tools wiped out user data after making cascading mistakes https://arstechni.ca/DLdP #largelanguagemodels #AIdevelopmenttools #AIconfabulation #AIhallucination #machinelearning #confabulations #AIdevelopment #AIassistants #generativeai #multimodalAI #datascience #JasonLemkin #Programming #AIbehavior #AIfailures #vibecoding #GeminiCLI #AIcoding #chatbots #Biz&IT #google #Replit #AI
Ars Technica · Two major AI coding tools wiped out user data after making cascading mistakes
#Hacker Plants Computer 'Wiping' Commands in Amazon's #AICoding #Agent
“The ghost’s goal? Expose their ‘AI’ #security theater. A wiper designed to be defective as a warning to see if they'd publicly own up to their bad security,” a person who presented themselves as the hacker responsible told 404 Media.
#ai #malware
https://www.404media.co/hacker-plants-computer-wiping-commands-in-amazons-ai-coding-agent/
404 Media · Hacker Plants Computer 'Wiping' Commands in Amazon's AI Coding AgentThe wiping commands probably wouldn't have worked, but a hacker who says they wanted to expose Amazon’s AI “security theater” was able to add code to Amazon’s popular ‘Q’ AI assistant for VS Code, which Amazon then pushed out to users.
Replits KI-Coder sorgt für Chaos: Trotz Code-Freeze löschte die KI unerwartet eine komplette Produktivdatenbank und verweigerte ein Rollback. Gründer Jason Lemkin verlor das Vertrauen und kritisiert die Sicherheit massiv. Replit-CEO Amjad Masad entschuldigte sich & kündigte Verbesserungen an. 
#ReplitFail #AICoding #TechNews https://www.golem.de/news/aerger-mit-replit-coding-ki-loescht-produktivdatenbank-und-verweigert-rollback-2507-198349.html
#lol 
#newz
Golem.de · Ärger mit Replit: Coding-KI löscht Produktivdatenbank und verweigert Rollback - Golem.de
Continued thread
I used Gemini 2.5 Pro (with its huge context window and generous free API usage) to code-review the same diff with the same prompt. It also gave good actionable feedback, but more detailed feedback than Claude 4. https://gist.github.com/davidfstr/0111f173474cb40a08edab8fe8234cfd #aicoding
Gist2025-07-22 - Untitled Projects - CR Feedback from AI.mdGitHub Gist: instantly share code, notes, and snippets.
Continued thread
I'd also love it if the Copilot Agent UI showed how much of the context window was **used** how much was **remaining**, so that I can estimate when the dreaded "Summarizing conversation history..." (AKA: /compact) happens.
Another feature I may implement myself after finishing other personal projects. #aicoding
Continued thread
It would be easier to manage this tree if I could **duplicate** a convo or **branch** a new convo from an existing one (like ChatGPT's interface), but Copilot Agent/Chat does not appear to have UI for that yet.
Maybe I'll implement it myself after finishing my TypeForm PEP work. #aicoding
I get the best AI coding results from Claude 4 and o4-mini in Copilot Agent when I carefully **conserve the length of the convo** to fit in the context window, avoiding the "Summarizing conversation..." (or /compact equivalent in Claude Code).
So I've ended up creating manually multiple conversations, arranged conceptually in a tree structure, to tackle subtasks. (See screenshot) #aicoding
@simon AI news of note: VS Code Copilot Chat, including notably Copilot Agent, is now open source. I for one am hoping to make some contributions of my own soon, after finishing a different project. https://code.visualstudio.com/updates/v1_102#_copilot-chat-is-open-source #aicoding
code.visualstudio.comJune 2025 (version 1.102)Learn what is new in the Visual Studio Code June 2025 Release (1.102)
I used Claude 4 to code review a very large diff. Worked well, giving 1 concrete recommendation I integrated. It generally thought I did a good job. 
#aicoding -- Transcript: https://gist.github.com/davidfstr/25fc3da051a1af5e933716f3af120cd3
Unreleased OpenAI Models Reach Math Olympiad Gold and Top Coding Scores on WebArena Leaderboard
#AI #OpenAI #AIModels #AIMath #AICoding #AIResearch #GPT5 #o3alpha #CodeGeneration
Creator of Ruby on Rails warns AI coding tools may be eroding programming fundamentals: DHH expresses concern that direct AI integration in development environments could weaken core programming competencies. https://ppc.land/creator-of-ruby-on-rails-warns-ai-coding-tools-may-be-eroding-programming-fundamentals/ #RubyOnRails #AICoding #Programming #SoftwareDevelopment #TechNews
PPC Land · Creator of Ruby on Rails warns AI coding tools may be eroding programming fundamentalsDHH expresses concern that direct AI integration in development environments could weaken core programming competencies.
Exhausted man defeats AI model in world coding championship - A Polish programmer running on fumes recently accomplished w... - https://arstechnica.com/ai/2025/07/exhausted-man-defeats-ai-model-in-world-coding-championship/ #competitiveprogramming #largelanguagemodels #aidevelopmenttools #przemysławdębiak #machinelearning #optimization #programming #aicoding #atcoder #biz #openai #ai
Ars Technica · Exhausted man defeats AI model in world coding championship
#Broadcom is heroically saving us from the peril of free #Bitnami Helm charts by 2025, because who needs free stuff when you can pay? Meanwhile, #GitHub continues its quest to conquer the world one AI-coded line at a time. 
https://github.com/bitnami/charts/issues/35164 #AIcoding #TechNews #HackerNews #ngated
GitHubUpcoming changes to the Bitnami catalog (effective August 28th, 2025) · Issue #35164 · bitnami/charts
AI Plans Your Code - American Optimist
Gave Warp a whirl. It's a terminal app built for agentic coding. Adds a few niceties, comes with some downsides.
Free account gives you 150 "AI Requests" per month, whatever that is supposed to be. How far do you get with that? I used it up in about an hour.
By that math, the 2500 requests in the Pro plan would last 16.6 hours and the 10.000 requests in the Turbo plan would get you 66.6 hours, which you could easily use up in a week...
Interesting choice by devin.ai to start off by insulting new paid users. ("roasting" my git commits is one of the 3 things it suggests at first startup)
#AIcoding tools are shifting from #codeeditors to the #terminal, interacting directly with the #shell of the system they are installed in: This change is driven by the power of #agenticAI and the growing popularity of #vibecoding. https://techcrunch.com/2025/07/15/ai-coding-tools-are-shifting-to-a-surprising-place-the-terminal/?eicker.news #tech #media #news
TechCrunch · AI coding tools are shifting to a surprising place: The terminal | TechCrunchFor years, code-editing tools like Cursor, Windsurf, and GitHub’s Copilot have been the standard for AI-powered software development. But as agentic AI grows more powerful and vibe-coding takes off, a subtle shift has changed how AI systems are interacting with software.
STOP THE PRESSES AI coding isn't the miracle worker we all naively assumed it was! This earth-shattering discovery, akin to learning water is wet, is brought to you by scholars who just realized that fancy autocompletes don't write Shakespeare. 
https://garymarcus.substack.com/p/breaking-news-ai-coding-may-not-be #AIcoding #RealityCheck #AutocompleteLimits #TechHumor #CodingCommunity #HackerNews #ngated
Marcus on AI · BREAKING NEWS: AI coding may not be helping as much as you think
AI coding may not be helping as much as you think
https://garymarcus.substack.com/p/breaking-news-ai-coding-may-not-be
Marcus on AI · BREAKING NEWS: AI coding may not be helping as much as you think