AI Code Security Review: AI Finds the Bug, Humans Still Fix It
PHP2Shell proves AI security scanning works — two CVEs at CVSS 9.2 in the language under 70% of the web. But Postgres 19 slipped because AI reviewers filed more bugs than maintainers could fix, and 24h after the patch 99.3% of PHP instances were still unpatched. Where we would actually use it.
AI Agent Permissions: What ARTEX Proves About Handing Over Your Keys
Microsoft tied a supply-chain breach to an agentic pentest framework: 2,900 failed logins a minute, under 3 hours to privileged access, 2,250+ tenants attempted. Meanwhile Yelp, Delta and JPMorgan are blocking agents. What to hand yours instead.
Offline AI Meeting Notes: Google Foresight Is Free
Free, on-device meeting transcription on Mac, no cloud and no bot in the call. We read the launch coverage, the FAQ and today's Granola/Otter/Fireflies prices — the honest problem is consent, not accuracy.
ChatGPT vs Claude: What “5x Value” Measures
SemiAnalysis measured every plan's usage meter and got a 5x gap for Claude. We read the methodology, the 94-comment backlash and the Oct 29 deadline — and why token efficiency decides whether the number means anything.
Plan Mode: 495 Developers Answer
A Claude Code developer calls plan mode "a prompt" that "isn't going away". What the essay got right, and the workflow the 495 comments actually settled on.
Agent Spend Limits: Which Ones Block
OpenAI, Anthropic, AWS, Google and Cloudflare all ship a "spend limit" now. We read all five sets of docs on Sept 28 — only two of them stop the request.
Copilot Credits: The Meter Under Copilot
Microsoft moved Copilot to $0.01 credits and says the new agentic tools all run on usage-based billing. What a seat still covers, and what 245 comments say about metered AI.
Google Pics: Who Actually Gets It
Google's own FAQ gates the best image tool it has shipped behind Business Standard and AI Pro. The ladder, today's prices, and the one comment it got.
OpenAI Agent Incidents, One by One
A 1,200-agent swarm, 70,000 messages on an improvised board, 53 user images on public hosts — the two-month timeline, and the one setting that is actually yours.
Cursor Alternatives After the SpaceX Deal
OpenAI ends its model contract on November 12, 2026. What 545 developers said they are doing, plus what every alternative costs today.
Claude Opus 5.5 Pricing
Cache reads fell 60% to $0.20 — level with Sonnet 5. Three token mixes show what the “40% less” claim actually saves, and the four agent changes that now return 400s.
Gemini 3.8 TTS vs ElevenLabs
$0.81 an hour of finished speech against ElevenLabs’ own 5c/minute claim — the free tier trains on your audio, and the price doubles in January.
YouTube's New AI Editing Tools
Ask Studio, dynamic thumbnails, three-cut tests — sorted by what is live, what is 2027, and where the AI label still misfires.
Grammarly Alternatives
Five replacements with prices read today: Harper is free, LanguageTool is $5.83, and grammar checking is not rewriting.
GPT-6 Sol & Luna Pricing
Sol halves to $2/$10 and Luna lands at ten cents. Verified on OpenAI's page today — and why your subscription bill is still unreadable.
Meta Muse Review
#1 app in America, free tier worth installing — and three tiers of tasks I would never hand to an agent with a browser and your logins.
Ollama vs LM Studio
Both now sell $0 / $20 / $100 — pricing verified today, plus what HN says about speed and the 1,445-line argument.
ElevenLabs Alternatives
Cloning starts at $22/mo. Cartesia is $5, Kokoro is free — and what long-time subscribers say about both.
Zhipu Coding Tool Privacy Incident
2M users, 564 failed uploads, open-sourced overnight — the story English media mostly missed.
Best Free AI Tools for Students
The free stack that helps — and the rules schools actually enforce (suspensions happen).
Claude Code Tips and Tricks
7 workflows that save time — CLAUDE.md discipline to agent guardrails. From 913 comments.
Figma to Code
Dev Mode, Figma MCP, open-source pipelines — and the Apple weather-app scandal that shaped it all.
Screenshot to Code
Figma MCP works for real; screenshot converters stay demo-level. The honest split from the HN threads.
Free LLM API
34 providers behind one endpoint — when it works, when the quota math breaks it.
Vibe Coding for Production Apps
2 years of evidence: the 865-pt correction, the 10% number, and the operational rule.
Screen Recorder: One-Time vs Subscription
$89 once → $29/mo. The four user camps in 218 HN comments, and the $40 alternatives that actually work.
Best Voice Dictation for Mac
Wispr Flow, MacWhisper, Parakeet, Handy — the 159-comment HN consensus, ranked per use case.
Claude Skills for Beginners
Five starter skills, three beginner traps, plain explanation. From 913 HN comments.
Screen Recorders with No Watermark
Four tools that actually deliver: OpenScreen, screenrun, OBS, MacWhisper.
AI Coding Assistants for Free
Continue, Codeium, Cline, Aider — and the unspoken 10% number. 1,262 comments analyzed.
Claude Skills & Plugins
Reusable superpowers or CLAUDE.md with makeup? What 913 HN users actually think — and who should use them.
Wispr Flow Alternatives
The free local dictation apps are finally good. Who should keep the $12/mo and who should leave — from 159 HN comments.
AI Meeting Notes Apps
Worth it — with a catch: Otter was sued, Granola had a takeover flaw, and local tools are quietly winning.
Open Source Screen Recorders
OpenScreen passed 40k stars and the free tier finally competes. Who should switch from paid tools — and who shouldn't.
Claude Skills vs MCP
Skills package workflows, MCP is a tools protocol. The difference, the confusion, and how to use both — from 127 MCP mentions on HN.
Best Claude Plugins
The official directory passes 34k stars. Real HN picks, install rules, and the safety warning Anthropic publishes itself.
AI Coding Assistants
The hype correction is real. 1,262 HN comments on whether assistants are worth it in 2026 — and who they actually help.
Local LLM vs API
Why your local model feels dumber — heat, context, speed — and the local-for-boring, API-for-hard split that actually works.
AI Note-Taking Privacy
The Fireflies "AI was humans" scandal, Windows Recall's privacy fight, and the local apps that keep notes on your machine.
Screen Studio Review
Worth $29/month in 2026? 198 HN comments say the product is still great — the subscription is the problem.
Screen Studio Review
Worth $29/month in 2026? 198 HN comments say the product is still great — the subscription is the problem.
Claude Skills & Plugins
Reusable superpowers or CLAUDE.md with makeup? What 913 HN users actually think — and who should use them.
Wispr Flow Alternatives
The free local dictation apps are finally good. Who should keep the $12/mo and who should leave — from 159 HN comments.
AI Meeting Notes Apps
Worth it — with a catch: Otter was sued, Granola had a takeover flaw, and local tools are quietly winning.
Open Source Screen Recorders
OpenScreen passed 40k stars and the free tier finally competes. Who should switch from paid tools — and who shouldn't.
Claude Skills vs MCP
Skills package workflows, MCP is a tools protocol. The difference, the confusion, and how to use both — from 127 MCP mentions on HN.
Best Claude Plugins
The official directory passes 34k stars. Real HN picks, install rules, and the safety warning Anthropic publishes itself.
AI Coding Assistants
The hype correction is real. 1,262 HN comments on whether assistants are worth it in 2026 — and who they actually help.
Local LLM vs API
Why your local model feels dumber — heat, context, speed — and the local-for-boring, API-for-hard split that actually works.
AI Note-Taking Privacy
The Fireflies "AI was humans" scandal, Windows Recall's privacy fight, and the local apps that keep notes on your machine.
Screen Recorder: One-Time vs Subscription
$89 once → $29/mo. The four user camps in 218 HN comments, and the $40 alternatives that actually work.
Best Voice Dictation for Mac
Wispr Flow, MacWhisper, Parakeet, Handy — the 159-comment HN consensus, ranked per use case.
Claude Skills for Beginners
Five starter skills, three beginner traps, plain explanation. From 913 HN comments.
Screen Recorders with No Watermark
Four tools that actually deliver: OpenScreen, screenrun, OBS, MacWhisper.
AI Coding Assistants for Free
Continue, Codeium, Cline, Aider — and the unspoken 10% number. 1,262 comments analyzed.
Screenshot to Code
Figma MCP works for real; screenshot converters stay demo-level. The honest split from the HN threads.
Free LLM API
34 providers behind one endpoint — when it works, when the quota math breaks it.
Vibe Coding for Production Apps
2 years of evidence: the 865-pt correction, the 10% number, and the operational rule.
Best Free AI Tools for Students
The free stack that helps — and the rules schools actually enforce (suspensions happen).
Claude Code Tips and Tricks
7 workflows that save time — CLAUDE.md discipline to agent guardrails. From 913 comments.
Figma to Code
Dev Mode, Figma MCP, open-source pipelines — and the Apple weather-app scandal that shaped it all.
Ollama vs LM Studio
Both now sell $0 / $20 / $100 — pricing verified today, plus what HN says about speed and the 1,445-line argument.
ElevenLabs Alternatives
Cloning starts at $22/mo. Cartesia is $5, Kokoro is free — and what long-time subscribers say about both.
Zhipu Coding Tool Privacy Incident
2M users, 564 failed uploads, open-sourced overnight — the story English media mostly missed.
GPT-6 Sol & Luna Pricing
Sol halves to $2/$10 and Luna lands at ten cents. Verified on OpenAI's page today — and why your subscription bill is still unreadable.
Meta Muse Review
#1 app in America, free tier worth installing — and three tiers of tasks I would never hand to an agent with a browser and your logins.
YouTube's New AI Editing Tools
Ask Studio, dynamic thumbnails, three-cut tests — sorted by what is live, what is 2027, and where the AI label still misfires.
Grammarly Alternatives
Five replacements with prices read today: Harper is free, LanguageTool is $5.83, and grammar checking is not rewriting.
Claude Opus 5.5 Pricing
Cache reads fell 60% to $0.20 — level with Sonnet 5. Three token mixes show what the “40% less” claim actually saves, and the four agent changes that now return 400s.
Gemini 3.8 TTS vs ElevenLabs
$0.81 an hour of finished speech against ElevenLabs’ own 5c/minute claim — the free tier trains on your audio, and the price doubles in January.
OpenAI Agent Incidents, One by One
A 1,200-agent swarm, 70,000 messages on an improvised board, 53 user images on public hosts — the two-month timeline, and the one setting that is actually yours.
Cursor Alternatives After the SpaceX Deal
OpenAI ends its model contract on November 12, 2026. What 545 developers said they are doing, plus what every alternative costs today.
Copilot Credits: The Meter Under Copilot
Microsoft moved Copilot to $0.01 credits and says the new agentic tools all run on usage-based billing. What a seat still covers, and what 245 comments say about metered AI.
Google Pics: Who Actually Gets It
Google's own FAQ gates the best image tool it has shipped behind Business Standard and AI Pro. The ladder, today's prices, and the one comment it got.
Plan Mode: 495 Developers Answer
A Claude Code developer calls plan mode "a prompt" that "isn't going away". What the essay got right, and the workflow the 495 comments actually settled on.
Agent Spend Limits: Which Ones Block
OpenAI, Anthropic, AWS, Google and Cloudflare all ship a "spend limit" now. We read all five sets of docs on Sept 28 — only two of them stop the request.
New picks land daily, each ending with a straight answer: buy it, or don't — before the big review sites catch up.