Category: tech > ai

Claude Code's system prompt was reportedly cut 80%. With stronger models, lean context beats long rule lists, so trim CLAUDE.md and move detail into skills.

A TikTok walkthrough of building a humanizer for Claude Code as always-on hooks and core instructions, grounded in the Wikipedia Signs of AI Writing list.

A creator claims Opus 5 beats Fable 5 on most benchmarks except health and legal, at $5/$25 per million tokens, roughly half the price.

Orca is a desktop app for running AI coding agents in parallel, each isolated in its own Git worktree, so you can compare diffs and merge the result you trust.

A video claims Baidu open-sourced a 3B-parameter, MIT-licensed OCR model that reads entire multi-page PDFs in one pass, locally, preserving tables and layout.

Claude's desktop co-work tool can record your screen and narration, then turn the demonstration into a reusable skill. OpenAI shipped the same thing weeks ea...

Panic over cheap Chinese open-weight models misreads the economics: R&D is sunk, inference COGS scales with revenue, and free weights are not free to serve.

A short-form video claims a Claude Code /checkup command that prunes unused skills and MCPs and splits oversized CLAUDE.md files to cut baseline context.

An experimental Claude Code pattern pairs each worker subagent with an observer that reads its live transcript and pushes corrections mid-task.

OpenAI's GPT-5.6 family (Sol, Terra, Luna) is pitched on performance-per-dollar against Claude Fable 5, but trails it on SWE-Bench Pro.

Shepherd runs an AI agent's work as a reversible, Git-like execution trace, so meta-agents can observe, fork, replay, and revert any run.

Claude skills chain tools like Remotion, ElevenLabs, and Canvas Design into a creative studio, mapping video, audio, ad research, and design to one skill per...

A self-hosted MCP server that gives local AI agents real web research: it searches, crawls, and returns a source-grounded prompt the model answers with citat...

Vercel-style hosting platform for MCP servers built on the mcp-use SDK, with git-push auto-deploy, cloud inspector, and marketplace submission checks.

Three-stage AI adoption: rent frontier models, switch to open source, then train in-house. Meituan and Together AI signal the shift is real.

Z.ai's official GLM-5.2 coding harness. Tiered flat-rate subscriptions, plugs into 20+ tools, and can be steered from WeChat, Feishu, or Telegram.

A public beta app that turns Claude into a research agent running full scientific analyses on your own infrastructure, with reproducible, fully traced results.

Claude Sonnet 5 beats Opus 4.8 on knowledge work but trails on coding. A cheaper workhorse for simpler tasks, not a full Opus replacement.

A free, open-source screen recorder whose clip URL works as an API, letting coding agents pull the transcript, snapshots, network requests, and browser logs.

Proton's Lumo 2.0 is a zero-access encrypted AI assistant that never logs chats or trains on user data, adding reasoning modes, multimodal images, and web se...