All updates in plain English

Copilot Fusion is now generally available: it mixes AI models for each prompt, and GitHub says it nearly matches Claude Code at half the cost.

Companies buying GitHub Enterprise through a sales team can now get Copilot included for every user; self-serve Copilot plans are unchanged.

Now generally available: describe repo chores like issue triage or test writing in plain Markdown, and GitHub runs a coding agent to do them.

The Copilot app's Projects preview lets you describe a goal and add repos and notes; Copilot runs agents and returns PRs. Automations can be shared.

GitHub open-sourced Replica (MIT): describe an idea in the Copilot app and get an interactive prototype built from your own design system.

Microsoft Foundry now offers the multi-model agent engine behind GitHub Copilot, as an SDK and managed service, for building your own long-running agents.

Vercel CI (beta) runs your tagged tasks on every push, only for code you changed; Vercel says its own wait for green builds fell 2-6x.

Save a model with your preferred settings under your own name in Vercel AI Gateway, then change it in one place without redeploying apps.

Supabase's public beta lets you pick the OrioleDB database engine on any plan, built for tables that update often. Not production-ready yet.

Qwen Code 0.26.3 adds a persistent Workflow mode and stops its /autofix-pr command from overwriting work you haven't pushed.

Cloudflare's Agents SDK lets an agent open a private dev server in a real browser without exposing it to the public internet.

Claude Code 2.1.312 stops it overwriting very large instruction files after reading only part of them, and finds folder-specific skills again.

Codex CLI 0.165.0 adds Google Cloud Vertex AI as a built-in provider and the GPT-6.1 Terra model, and lets you change reasoning effort mid-chat.

Replit's agent can now edit your team's existing codebase from a plain-language request, test it in a sandbox, and open a PR for review.

Supabase Compute (private alpha, waitlist) lets you run AI agents, background jobs and APIs next to your database; us-west-2 only for now.

Vercel now deploys a root Dockerfile or Containerfile as a container when no framework is detected, and sends all traffic to it.

Vercel Agent sessions now cost 67% less on average after a rebuild of how it manages context; there's nothing for you to do.

Services built on the Copilot SDK can now use Copilot as a GitHub App on your organization, with no signed-in user needed (public preview).

A new beta command-line tool, gemini-api, lets coding agents, scripts and CI generate text and media and run Gemini agents, with no dependencies.

CopilotKit 1.81.0 adds built-in generative UI support plus server-side additions to its runtime; nothing breaks.

OpenAI's GPT-6.1 Terra is a fast, cheaper model with a 1M+ token context; API prices are 50% lower than GPT-5.6 Terra and GPT-6.1 Sol.

Copilot now has a model review each command, file or URL request; clearly safe ones go ahead and the rest still ask you.

Claude Code 2.1.311 is a big fix release: long sessions now compact and keep going instead of stopping with a credits error.

Vercel Agent can now check each monitor alert first and send only the ones it rates Critical to your paging system.

You and your agents can now search Vercel Connect's 1,000+ services from the terminal by name, keyword or category.

Cloudflare Computer 0.7.0 lets agent code resume after a deploy or crash and pause for approvals; replayed calls can repeat, so mark unsafe ones.

New paid-tier Gemini API projects now save Generate Content requests by default; existing projects keep their setting, and you can opt out per request.

Qwen Code 0.26.1 adds an /autofix-pr command that pushes fixes and re-runs checks, and lets scheduled tasks wait on a shell check.

The Chroma vector database team is joining Replit. Open-source Chroma continues, but hosted Chroma Cloud winds down within 60 days, so plan a move.

The Vercel CLI's new vercel budgets command shows team spend, sets or removes a spending budget, and can optionally pause production deployments at 100%.

Vercel Connect can now create a separate restricted Stripe API key per agent or task, store it for you, and rotate it without redeploying.

Codex CLI 0.164.0 adds ten saved-prompt slots (/presets), quicker /model and /permissions switching, and hides credentials in supported remote shell sessions.

Replit Agent can suggest new workspace skills for your approval, and eligible top-up customers' published apps now keep running when credits run out.

Supabase launched a public-beta Knowledge Base at supabase.com/kb with troubleshooting and migration guides for developers and their coding agents.

Moonshot's Kimi K3.1 Pro is a new open-weight model with a 1M-token context for coding and agent work; full weights are due by Nov 4.

Claude Code 2.1.310 brings the Monitor tool to Bedrock, Vertex and Foundry, and fixes cleanup deleting recently pasted images.

GitHub's Teams app lets you @mention GitHub on a bug so Copilot investigates and opens a fix; the old app retires Nov 16.

Anthropic gave the Agent Skills standard, used by 70+ agent platforms, to the Linux Foundation; it stays free to use without joining.

Run vercel deploy --temporary to get a live link for a prototype before signing in; claim the project within an hour to keep it online.

You can now reach the Claude API, run by Anthropic, through your Google Cloud project, with Google Cloud access controls and billing.

OpenAI's Decisions API, for structured answers like classify or route, is out of beta at $0.10 per million input tokens with no output charge.

Replit Monitoring Agent reads your published app's logs, investigates alerts, validates a fix and emails a summary; it can auto-deploy fixes if you opt in.

Claude Code 2.1.309 is mostly security fixes: sandboxed shell commands no longer run startup files outside the sandbox.

Google open-sourced a design-first plugin for Antigravity: one agent drafts a design doc, then fresh agents review it before any code is written.

Cloudflare's MCP servers now only work at /mcp; old /sse links return an error, so update your configs and sign in again.

Cloudflare Agents SDK 0.30.0 lets agents sleep and recover from storage errors more gracefully, but you can't roll back past this upgrade.

Vercel Labs released fx, an experimental open-source coding agent written in Zig, built to be scripted or embedded rather than used full-screen.