ai. 189 essays.

Cloudflare began continuously checking whether verified bots honor robots.txt and rate limits, Google rewrote its Indexing API quota pages, and two WordPress plugin advisories landed.

Claude Code 2.1.302 enables fork subagents by default, OpenCode 1.19.3 generates reasoning variants, Codex 0.164 alphas appear, and GitHub flags Copilot model retirements.

Google reshaped how AI answers hand traffic back to the web on October 15, Universal Cart landed in US Search, and Chrome 155 hit stable with JPEG XL. Ranking systems stayed quiet.

Claude Code 2.1.301 stops unattended retries from waiting on far-off usage limits, Codex 0.163 lets you steer a response mid-flight, Copilot CLI reaches 1.0.98, and OpenCode 1.19.2 keeps 1.x installs on 1.x.

Claude Dashboards (beta) builds live dashboards on top of your warehouse. Here is an advanced DevOps one: SLO error budgets with 1h and 6h burn rates, the 5xx rate with deploy markers, the deploys most likely behind a spike, and DORA, with every query and the setup.

Cloudflare shipped an emergency WAF rule that blocks by default, Rank Math patched an unauthenticated data wipe, and ChatGPT's outbound links appear to have gone untagged.

Two weeks after a testers-only launch, Google opened Gemini 4 Argon to paid developers, businesses and Gemini Pro and Ultra subscribers on October 13. Here is where you can use it, what the intro price means after November 30, how it scores on independent tests, and how to make a first API call.

Bing Webmaster Tools now clusters AI-search citations by theme and names the queries where rivals get cited instead of you. Cloudflare RUM adds LCP and INP phase data.

Claude Code adds idle compaction and Haiku 5.5, GitHub Copilot ships Gemini 4 Argon and GA local sandboxing, Codex 0.162 adds managed worktrees, and OpenCode 1.19 points to v2.

Most help desks bill you per agent, every month. Helpdock is the open-source alternative I am building: email, WhatsApp, Instagram, Telegram and live chat in one inbox, SLAs and rules, AI that answers from your own docs, a help center, many brands per install and full Arabic support.

Meta released Muse Wonder and Muse Wonder Full on October 12, its first frontier models. They focus on memory across conversations and agents that keep going when blocked. They rank close to the top but not first, cost the same as Claude Opus 5.5, and have one data setting to check.

Codex 0.162 and OpenCode 2.0.27 both ship managed Git worktree tools, VS Code 1.141 shows Copilot sessions side by side, and Claude Code 2.1.295 adds opt-in fail-closed hooks.

DevOps decisions get lost across tickets, docs and chat. Here is how to keep ADRs, incidents and runbooks in one Obsidian vault, track them with the new Bases Kanban view, let an AI agent answer from them, and sync it all through your own server.

Local sandboxing for GitHub Copilot is now generally available, OpenCode 1.19 stops auto-updating, and Claude Code and Codex ship fresh point releases. Here is what matters.

Claude Code 2.1.295 makes failing hooks block actions, Codex CLI 0.162 adds managed Git worktrees alongside GPT-6.1 Sol Ultrafast, and GitHub makes stacked pull requests generally available.

Claude Code 2.1.293 and 2.1.294 add Claude Haiku 5.5 and fix hooks that failed to block actions. GitHub Copilot makes local sandboxing generally available, adds Haiku 5.5, and can now discover Ollama models in the CLI.

Claude Code 2.1.292 lets you set effort per sub-agent and tightens security, Copilot CLI makes command sandboxing available to everyone, and Copilot code review gains an API.

Claude Code ships managed agents onboarding and a quick regression fix, Copilot CLI adds a local/cloud environment picker, and Codex 0.160.1 fixes Windows MCP launches.

What moved over the weekend: fresh rules for reviewing machine-written drafts, an unfinished spam rollout, and two upgrades worth a same-day pull request.

OpenAI DevDay brought reusable cloud environments, a voice-driven CLI and Codex Security Cloud, while GitHub Copilot added GPT-6.1 Sol, a code review API and retired four models.

Claude Code 2.1.289 closes several gaps where deny and ask permission rules were not enforced, and Copilot CLI 1.0.92 adds an environment picker and sandbox prompts. Codex remains in alpha builds.

GitHub Copilot can now drive desktop apps and run dynamic multi-agent workflows, four Copilot models were retired, and Claude Code 2.1.288 adds mod selection support and built-in gh api in cloud sessions.

A dated, source-checked look at what moved in search over the past day — split into an editorial track and an engineering track, with the unverified items flagged as such.

Claude Code 2.1.287 introduces Mods and a built-in 'You should know' side agent, while Codex CLI 0.160.0 adds an agent command center and Guardian review options.

GitHub Copilot's HydraFusion research preview orchestrates multiple models per task, Claude Code 2.1.286 fixes resume and auth bugs, and OpenCode 1.18.34 and Codex CLI 0.159.3 ship smaller updates.

What verifiably changed in search on 29-30 September 2026, split into an editorial track and an engineering track, with every claim traced to a primary source.

GPT-6.1 Sol is now the default in Codex CLI and generally available in GitHub Copilot, Codex CLI 0.159 adds instant interrupt, and Claude Code 2.1.285 adds plugin configuration and a WebFetch kill switch.

Google quietly opened AI Mode's monitoring agent to everyone, the spam update is only halfway through, and Cloudflare shipped a gentler alternative to purging your cache.

Claude Code v2.1.284 starts sessions in auto mode by default and ships Sonnet 5.5 and Opus 5.5 as the default models, while GitHub Copilot rolls out Sonnet 5.5 across its IDEs, CLI and mobile apps.