Quick one this week: a big open-weights model, a standards handoff, and Anthropic showing off how fast code is now being shipped. Here's what I'd actually pay attention to if you build with AI daily.

This Week in AI Coding

• Kimi K3.1 Pro lands on the API. Moonshot prices it at $2 / $10 per million tokens, and says weights follow by Nov 4. Their own benchmarks put it near Opus 5.5 and GPT-6 Astra; Artificial Analysis scored it 55 vs 58 for Opus 5.5. Worth a test as a cheaper fallback model, but wait for the weights before you plan around "open". (Simon Willison)
• Anthropic donates the Agent Skills standard to the Agentic AI Foundation (Linux Foundation), next to MCP. If you've been writing skills, you're now writing against a vendor-neutral spec, which is the good outcome. (Simon Willison)
• Anthropic replaced Git internally with a system called Scion. The reason: CI checkouts would have meant over 600 PB of git traffic a day, and merged changes went from about 2,500/day in May to over 30,000/day by September. That is what agent-scale code volume does to your tooling. (Simon Willison)
• Claude Mods: plugins as "middleware for the agent harness". TypeScript/JS functions that hook into prompts, tool calls and rendering. Powerful, but they run with your permissions and aren't sandboxed, so read a mod before you install it. (Latent Space)
• Codex CLI 0.164.0 adds ten /presets prompt slots, masks credentials in remote shell sessions, and fixes failed compaction leaving partial summaries in history. Boring, useful. (OpenAI Codex release)
• Anthropic told investors Q3 revenue rose nearly 50% to $16.8bn ahead of an IPO, and that distillation by rival labs was 5-7% of revenue before they cut it. Reported figures, not audited ones. (Simon Willison)

Tool of the Week

Claude Code v2.1.310 (shipped Oct 23). Nothing flashy: the Monitor tool now works on Bedrock, Vertex and Foundry, and there are fixes for compaction after a "Prompt is too long" error and for background cleanup deleting recent files such as pasted images. If you run long sessions, update. (release notes)

Quick Links

• My obituary for programming - Simon Willison, worth reading with a coffee
• Battle Drones: a vibe coded RTS with audio-based multiplayer
• AI Is Getting Good at Math. Here's Why. - Latent Space
• AINews: Kimi 3.1 Pro

הדופק / The Pulse

What changed this week

• [Models] Kimi K3.1 Pro on the Kimi API, $2/$10 per million tokens, weights promised by Nov 4. Hype / unverified: the "near Opus 5.5" claim is Moonshot's own benchmarks; the 55 vs 58 Artificial Analysis score is third-party. Source: https://simonwillison.net/2026/Oct/23/kimi-k31-pro/ (2026-10-23)
• [Harness] Agent Skills standard donated to the Agentic AI Foundation. Verified (Anthropic announcement as reported by Simon Willison): https://simonwillison.net/2026/Oct/22/donating-agent-skills/ (2026-10-22)
• [Harness] Claude Code v2.1.310. Verified (official release notes): https://github.com/anthropics/claude-code/releases/tag/v2.1.310 (2026-10-23)
• [Harness] Codex CLI 0.164.0. Verified (official release notes): https://github.com/openai/codex/releases/tag/rust-v0.164.0 (2026-10-23)

AI jobs this week

• Member of Technical Staff (AI Software Engineer, Agents) - Perplexity (San Francisco, onsite) - $220,000-$405,000 listed
• Member of Technical Staff (Software Engineer, Training Infrastructure) - Perplexity (San Francisco, onsite) - $220,000-$405,000 listed
• Software Engineer - Storage - Supabase (Remote)
• Staff Software Engineer, Payments Intelligence - Stripe (Seattle, New York, San Francisco, onsite)

337 new roles this week, 5,533 tracked (LLMHire API, 2026-10-25). More at llmhire.com.

The Pulse, all week: https://www.vibe-coding.academy/he/pulse/

Start here: the free first lesson at https://www.vibe-coding.academy/he/start/ - or the Builder plan at https://www.vibe-coding.academy/he/pricing/ ($19, about ₪70, 14-day refund).

endofcoding.com · vibecodingebook.com · llmhire.com · Vibe Coding Academy · YouTube @thevibecodingacademy