Coding agents & orchestration

• Claude Code v2.1.234 released: adds CLAUDE_CODE_PROJECT_DIR_NAME for per-project transcript dirs on multi-tenant hosts, plus a selection:clear keybinding.
• Orca is a new ADE for running Claude Code, Codex, OpenCode, or Pi side by side, each in its own git worktree, with a shared quick-open and usage/rate-limit tracker across accounts.
• Cursor's $60B acquisition by SpaceXai closes: AI News recap of the deal plus Cursor's arc from 5-person team to enterprise FDE player.
• GitHub: canvases for agentic workflows: argues chat loses agent work in scrollback, canvases keep multi-step agent state visible and steerable.
• Claude to Claude message visualizer (Reddit): tool for tracing agent-to-agent messages across sessions/machines, relevant if you're debugging swarm/fleet setups.
• Terminal picker for Claude Code sessions (Reddit): one picker across all running CC sessions, same problem your orphaned-swarm-session clutter hits.
• Prompt-only pane showing just prompts/replies (Reddit): strips tool-output noise from the CC terminal view, worth a look if scrollback fatigue is a problem for you.

Agent memory & cross-tool handoff

• ai-memory: long-term memory layer letting you quit Claude Code mid-task and resume in Codex without re-explaining architecture/failed approaches, cross-platform (Linux Docker/CI first).
• TencentDB Agent Memory: team-level memory hub turning chats/docs/code into four shared asset types (Chat Memory, Skill, LLM-Wiki, Code-Graph) across agents/frameworks.
• Open-sourced tool for switching AI coding tools (Reddit): same handoff problem as ai-memory, author wants feedback.
• octopodas: beating mem0 on long-eval memory: claims to beat mem0 on long-horizon agent memory eval, author's take is blunt dismissal of the benchmark's relevance.
• SnapState: persistent state for agent workflows (HN): checkpoint/resume state store for multi-step agent runs.

Model releases & LLM mechanics

• GLM-5.2 is a step change for open agents: Nathan Lambert flags a capability threshold crossed for open-weight agentic models, plus a new paper on open RL recipes for terminal agents.
• Qwen 3.8 27B overthinks everything: Apache-2.0, vision-capable, laptop-sized model; Willison's main gripe is excessive reasoning overhead on simple tasks.
• Claude Fable 5 and new AI safety fables: analysis of Anthropic's Fable 5 release through the lens of frontier-lab safety politics, notes a classifier fix to model manipulation of AI-research queries.
• Ontologies Are So Back: argues ontologies are being revived to keep probabilistic agents inside deterministic guardrails, talk from AI Engineer World's Fair.

CLI/TUI tooling

• llmfit: one command to check which models/providers actually run on your hardware, now with a benchmark-and-share mode contributing real tok/s numbers back via PRs.
• omlx: Apple Silicon LLM inference server with continuous batching and tiered SSD/KV caching, managed from the macOS menu bar.

Agentic SaaS & infra

• Stripe buys OpenRouter for $7B: ~50x on OpenRouter's reported $140M ARR, 90 days after its $1.3B Series B, signals infra/distribution consolidation over raw model plays.
• TinyFish: pitched as a "web operating layer" for AI agents, browsing/interaction infra.
• Clears: positions itself past "AI coding" into full agentic software delivery.
• Chert: Vapi-style AI video agents for FaceTime, low-code voice/video agent embedding.

Misc dev tools

• addyosmani/agent-skills: packaged senior-engineer skills (define/plan/build/verify/review/ship) for coding agents to follow consistently.
• OpenTrade: open-source trading harness built for Claude Code/Codex.
• Blender Agent Bridge: open-source MCP bridge connecting Blender to AI agent workflows.

next: confirm where to write this in vault, then commit.