Hands-on experience with AI, self-hosting, Linux, and the developer tools I actually use

Latest

Hands-on experience with AI, self-hosting, Linux, and the developer tools I actually use

Cheap AI Tokens Are a Scam Where Your Prompts Are the Product

Cheap AI API resellers undercut official prices by 70 to 97 percent because the discount is not the product: your prompts are. They log every request to resell as training data, route you to weaker models, and run on stolen-card accounts. A CISPA Helmholtz audit caught silent model swapping, but the harvested logs are the real margin.

Key Takeaways

  • A 90 percent discount on frontier AI is funded by reselling your prompts.
  • Proxies can send an “Opus” request to a cheaper model and relabel it.
  • Many reseller accounts come from stolen cards and faked identity checks.
  • Pointing a coding agent at an unknown API host hands a stranger your machine.
  • Official APIs and zero-retention gateways are cheap enough to skip the scam.

Why is a Claude or GPT API 90% cheaper from a reseller?

A frontier model has a hard cost floor. GPU time per token is a real expense, and the official provider already prices it close to the bone. So a reseller charging one tenth of that loses money on every call, unless something else pays the bill. The discount cannot come from being smarter about compute.

Pinterest's MCP Deployment: 66,000 Monthly Invocations and 7,000 Engineering Hours Saved

Pinterest’s Model Context Protocol rollout hits 66,000 calls per month across 844 active users. It’s the most detailed public case study of MCP at scale. A central registry, two-layer auth, safety reviews, and human checkpoints set this apart from a prototype. The payoff: about 7,000 engineering hours saved each month.

The story comes from Pinterest’s engineering blog post in March 2026 and later coverage by InfoQ . For any team weighing MCP for live use, this rollout is a solid guide.

Home Assistant Voice Preview Edition Review: Is the 59ドル Box Ready for Daily Use?

After more than a year of daily use, the Home Assistant Voice Preview Edition is ready for daily use, with caveats. It is the only 59ドル smart speaker on the market with zero cloud dependency, and for anyone who already runs Home Assistant it slots into existing automations with almost no friction. On the plus side you get fully local wake word detection, sub-second response on common commands, a capable far-field mic array, and a privacy story Alexa and Google cannot touch. The frustrations have been equally consistent: wake word accuracy drops in noisy rooms, the built-in speaker is too quiet for a kitchen, custom wake words require a training pipeline most users will not bother with, and anything beyond “turn the lights on” still needs either a local LLM or a cloud model piped through Assist.

Claude Code Remote Agents: Dispatch, Scheduled Tasks, and /loop Explained

Claude Code now ships four ways to run agents remotely: Dispatch, Remote Control, Scheduled Tasks, and /loop. Pick the wrong one and you either over-build a simple polling job or under-build something that needs real persistence. Each works at a different layer of the stack. Each has its own lifecycle, infrastructure needs, and rules for what survives a closed terminal or a sleeping laptop.

Dispatch: Send Tasks from Your Phone to Your Desktop

Dispatch launched on March 17, 2026 as a research preview inside Claude Cowork. Open the Claude mobile app, describe a task, and Dispatch routes it to your Claude Desktop instance on your dev machine. Claude Code runs the task locally with your file system, MCP servers, skills, connectors, and any other tools you’ve set up. The result comes back to your phone.

Gemini CLI Is Dead: Migrating to Antigravity CLI in 2026

Google shut down Gemini CLI on June 18, 2026, pushing free, Pro, and Ultra users onto the closed-source, Go-based Antigravity CLI . You can keep the open-source Gemini CLI running with a paid API key, but its separate quota pool is gone. Enterprise and Code Assist Standard licenses still work unchanged.

Key Takeaways

  • Gemini CLI stopped serving free, Pro, and Ultra accounts on June 18, 2026.
  • Its replacement, Antigravity CLI, is a closed-source Go binary, not open TypeScript.
  • You can still run Gemini CLI by feeding it a paid Gemini API key.
  • The catch: agy shares one usage pool, so quotas drain much faster.
  • Enterprise and Code Assist Standard licenses keep working unchanged.

What happened to Gemini CLI?

On June 18, 2026, Gemini CLI and the Gemini Code Assist IDE extensions stopped serving requests for free, Google AI Pro, and Ultra personal accounts. The Google Developers Blog announcement confirmed the cutoff and named the replacement: Antigravity CLI, run with the command agy.

COSMIC Desktop 1.0: One Month of Daily Driving System76's Rust DE

Thirty days of COSMIC Desktop 1.0 on Pop!_OS 24.04 LTS, and I am keeping it. Switch if you are a keyboard-first developer who wants a real tiling-plus-floating hybrid, appreciates Rust-grade stability, and likes a UI that sits under 900 MB at idle. Wait a release or two if you depend on a big GNOME extension collection, niche input methods (CJK/IBus edge cases), or a heavy accessibility stack. The COSMIC Store’s catalog is still smaller than Flathub’s GNOME Circle or the KDE offerings, and only a handful of third-party cosmic-ext-* applets exist in 2026. Everyone else should at least boot the live ISO before deciding. COSMIC 1.0 is the first new Linux desktop in a decade that does not feel like a fork of something older.

AltStyle によって変換されたページ (->オリジナル) /