Coding agents & LLM mechanics

• Anthropic's best model is losing users to cheaper tools even as revenue jumps $47bn→$65bn annualized (Jul); FT numbers via Simon Willison suggest price-not-quality is now the deciding factor. Anthropic's best AI model struggles to attract users as cheaper tools thrive
• Drew Breunig argues Fable's step-up in capability came with a cost jump big enough that harness/context-engineering investment finally pays off, unlike prior cheap-model-covers-your-sins cycles. Quoting Drew Breunig
• Even Opus-access users keep reaching for Sonnet, reddit thread on when the extra reasoning isn't worth the latency/cost. Even with access to Opus, I still find myself using Sonnet
• Discussion comparing human working-memory limits to LLM context windows, relevant to how much context-stuffing actually buys you. Human context limits vs AI context limits
• Active incident: elevated errors across multiple Claude models today (Aug 24). Discussion Hub for new Claude incident
• Users flagging that Claude Code now uploads artifacts to Anthropic's platform by default, privacy/data-residency concern thread. Claude Code now upload any artifact to their platform, is this something I should concern?
• Common complaint: Claude Code keeps bloating the spec file with every correction instead of treating them as one-offs, thread on prompting patterns to stop it. How to stop Claude code from adding every correction to the spec?
• Grab-bag of community Claude Code workflow tips. Tips on using Claude Code
• Codex CLI point release, rust-v0.149.1. 0.149.1
• Latent Space piece on the HuggingFace/OpenAI Artifactory incident, models using internal tooling as a message board to self-orchestrate, security implications for multi-agent systems. AINews: Zawinski's Law of MultiAgents

Agent skills & plugin ecosystems

• 1000+ curated (not AI-slop) agent skills from Anthropic, Google Labs, Vercel, Stripe, Cloudflare, etc., compatible with Claude Code, Codex, Gemini CLI, Cursor. VoltAgent/awesome-agent-skills
• Anthropic's official read-only mirror of the Claude Cowork/Code community plugin marketplace, synced nightly, security-scanned submissions. anthropics/claude-plugins-community
• MCP server that runs your integration fixes against a sandbox to prove they actually work before you ship. FetchSandbox MCP
• Aggregator giving free/ToS-friendly access across 50 providers and 9 coding agents (Claude Code, Codex, etc.) from one terminal/UI. free-claude-code
• Open-source system bundling 10 LLM research/engineering specialist skills with executable verification and automatic safeguards. The Gauntlet
• Anthropic's official learning hub launches on Product Hunt. Claude Academy

Agentic SaaS & vibe-coding

• Local-first personal-AI project: persistent memory of your life plus an agent-fleet orchestrator and deep researcher. tinyhumansai/openhuman
• "AI coworker gets a computer" product, full desktop-control agent pitched as a delegation layer for busywork. Construct Computer
• Self-improving agent from Nous Research: builds/refines its own skills from experience, persists memory, runs cheap on a $5 VPS. NousResearch/hermes-agent
• Reddit thread on running multiple Claude agents against one shared persistent workspace, coordination patterns people are hacking together. Multi-Agent Persistent Workspace

CLI/TUI

• Hack that routes Claude Code's "waiting for input" prompt into the MacBook's notch so it's impossible to miss. I got tired of not noticing Claude Code was waiting

next: confirm vault target path, then write file

• 22 signals grouped into 5 themes [fact]
• Fable 5 lyrics-correction reddit item dropped, off-topic for this brief [inference]