Coding agents & CLI

• Codex: 0 to 10M Users: OpenAI's Akshay Nathan on Codex MAU up >10x since Jan, thesis being the ~100x of people who use code but can't write it is the real prize once agentic interfaces get good enough.
• Claude Code v2.1.229: adds remote-control --continue for resuming sessions, server-supplied hooks for self-hosted runners, and SSE keepalives to stop long-thinking streams idle-timing-out.
• oh-my-pi: Rust-core (~80k LOC) terminal coding agent forked from Pi, 60+ providers, 31 tools, LSP+DAP integration, hash-anchored edits; PRs open to all as trial.
• Agent Plugins 1.0: cross-client plugin spec (AWS, Anysphere, Microsoft, OpenAI, Vercel) so one plugin runs in VS Code, Copilot CLI, and the Copilot app, chipping at the walled-garden agent ecosystem.
• addyosmani/agent-skills: packaged senior-engineer workflows (define/plan/build/verify/review/ship) as reusable skills for coding agents to follow consistently.
• GLM-5.2 is the step change for open agents: Nathan Lambert flags GLM-5.2 crossing a capability threshold for open-weight agentic models, alongside new open RL recipes for terminal agents.
• Your contributors are AI-first now: AutoGPT maintainer's playbook for repo instructions, gates, and boundaries to keep AI PR contributors in check.
• openwiki: CLI that has an agent read your codebase and generate/maintain a linked Markdown wiki, built as memory for agents with a visualizer for humans.
• Phind (Show HN): GPT-4-powered dev search; HN notes it already owns the niche, real problem is monetizing unlimited API calls.

LLM mechanics

• How to steal a Reasoning Trace: AINews on a paper showing reasoning-trace extraction is essentially speculative decoding in disguise, feeding the interpretability/alignment/CoT-monitoring debate.
• I beat mem0 on long eval memory: builder claims to outperform mem0 on long-horizon agent memory evals, relevant if you're picking a memory layer for multi-agent systems.
• "I gave Claude one MCP server...": single MCP server used to chain an LLM, image, and video model in one Claude run, illustrating MCP as a composition layer not just a tool call.

Vibe-coding & agent orchestration

• Orca: desktop/mobile ADE that runs Codex, Claude Code, OpenCode, or Pi side-by-side in separate git worktrees with unified search, account switching, and usage tracking across a fleet of agents.
• SnapState: persistent state layer for AI agent workflows, targets the "agent forgets everything between runs" problem.
• book-to-skill: converts a technical book/PDF into a Claude Code skill, claims 24-51x fewer tokens than dumping the book into context per query.
• agency-agents: native desktop app that browses a roster of specialized personality-driven agents and one-click installs them into Claude Code, Cursor, Codex, Gemini, etc.
• Ballet: describes-the-outcome workflow automation that writes real API integrations; HN pushback is it's a commodity market and "just write Python + Claude" already beats it.
• LaraCopilot: agentic engineer that builds full apps, another entrant in the vibe-coding-as-product wave.

Agentic SaaS & funding

• Lovable raises $400M Series C: $13.3B valuation led by Menlo/Scaleup Europe; community skeptical post-Claude Code and flags Lovable's generated code as bloated with unnecessary layers.
• OmniRoute: free MIT AI gateway aggregating 290+ providers/500+ models with quota-aware fallback and prompt-compression claiming 15-95% token savings, works with Claude Code/Codex/Cursor/Copilot.
• TencentDB Agent Memory: team-level shared memory hub turning chats/docs/code into governed memory assets (Chat Memory, Skill, LLM-Wiki, Code-Graph) across agents/frameworks.
• Cohesor: pitches itself as a neutral control plane for enterprise AI agents.
• Argos: lets AI agents attach screenshots/videos directly to pull requests.
• Grok Bot: xAI's "AI teammates you can give real work to" launch on Product Hunt.