A Unified Experience for All Coding Agents (code.visualstudio.com)

GitHub announced a unified agent experience in VS Code that brings multiple coding agents — including GitHub Copilot (cloud and CLI) and now OpenAI’s Codex on GPT-5/GPT-5 Codex — under one roof. OpenAI Codex is available to Copilot Pro+ subscribers with no extra account; Copilot mediates model calls and enforces standard rate limits. The goal is to reduce fragmentation (multiple subscriptions, tool-juggling, FOMO) by making VS Code “mission control” for agents so developers can pick the best agent for each task without leaving the editor. Technically, the release adds an Agent Sessions sidebar to orchestrate local and cloud agents, a tabbed “chat editors” UX that lets you watch progress and course-correct mid-run, and one-click delegation between agents. “Chat modes” are renamed to agents and developers get a built-in Plan agent that asks clarifying questions, recommends libraries, and can hand off full plans into the editor. For context management there’s runSubagent: spawn isolated subagents with their own context (via #runSubagent) to run side-research or deep dives and return only final results to the main chat. Custom agents are configurable and portable across Copilot CLI and other agents, and different models (e.g., Claude vs. GPT-5) can be tried for different strengths — improving prompt engineering, context control, and multi-agent workflows.