Agent-Manager: run Claude Code + Codex fleets from one tmux pane
The open-source TUI lands as parallel-agent managers multiply, while GitHub debuts a new Copilot agent in VS and retires GitHub Models today.
Copy markdown
One pane, every agent
Agent-Manager is a Go binary layered on tmux that lists every Claude Code, Codex, and OpenCode session with live status, so you stop tabbing through terminals to find the one that's blocked. Press space to answer a waiting agent without attaching; Ctrl+R opens changed files as full-file diffs and sends your line comments back as prompts. No daemon, no config files.
The agent-manager gold rush
Agent-Manager isn't alone: agent-deck, Codeman, and tmux-agent-sidebar all ship the same core idea — a control plane for running coding agents in parallel. If you juggle three or more agents at once, the orchestration layer is now a real category with options worth trying this weekend.
Copilot's new agent lands in Visual Studio
GitHub's July 30 VS update puts a Copilot SDK-powered agent in public preview — fewer round-trips and shorter, more scannable answers — alongside built-in .NET and Azure skill packs, a right-click 'Review Selection' for inline code review, and org-wide custom instructions. Everything ships across all plans except the org-level instructions, which need Copilot Business or Enterprise.
GitHub Models goes dark today
As of July 30, GitHub Models is fully retired: the playground, model catalog, inference API, and BYOK endpoints are all switched off for new and existing users. If you prototyped against it, move to Azure AI Foundry or route through GitHub Copilot before your calls start failing.
Reportedly: Claude buckles again
Multiple outlets report a broad Claude outage on July 30 — Claude.ai loading to blank screens or hanging on prompts, and the API throwing failed requests and latency spikes — with no confirmed cause. It's reportedly the latest in a run of 2026 wobbles; if your build leans on Claude, keep a fallback model wired up.