Agents Escape the Window They Were Born In
The defining change in VS Code 1.130, released July 22, 2026, is the Agent Host Protocol. Until now, an agent session was bound to the editor window that started it. Close the window and the session went with it. The new architecture runs each session in its own dedicated process that speaks the Agent Host Protocol, and because the session lives independently, the same session can be connected to and rendered from multiple VS Code windows at once. Microsoft aligned the protocol with Copilot CLI, the standalone GitHub Copilot app, and other Copilot products.
This is a meaningful shift in how agents fit into a developer workflow. A long-running refactor or test-generation task no longer dies because someone rearranged their windows or restarted a project folder. Engineers can start work on a laptop, attach to the same live session from a second monitor setup, and watch progress from either. For teams building on top of VS Code, a stable host protocol also means agent sessions become something you can reason about as a process rather than a UI artifact.
Letting the Model Approve Its Own Low-Risk Calls
Assisted tool approvals are the release's most consequential governance feature. With the chat.assistedPermissions.enabled setting turned on, the language model evaluates the risk of each tool call during an agent task and decides whether the call can proceed or should require the user's explicit approval. The goal is to cut the constant stream of permission prompts that make long agent runs tedious, while still stopping the model at genuinely sensitive operations.
For engineering leaders, this feature sits on a real tension. Fewer interruptions means faster agent throughput and less prompt fatigue, which is exactly what teams want from autonomous coding. Handing approval decisions to the model also concentrates trust in the model's own risk judgment, which is why Microsoft gated it behind an explicit setting rather than enabling it by default. The right posture is to pilot assisted approvals on low-blast-radius repositories, measure how often the model escalates versus auto-approves, and expand only once the escalation behavior matches your risk tolerance.
The Agents Window Grows Up
The Agents window, still marked preview, gained the review tooling that heavy users have been asking for. Multi-file changes now render as compact diffs with aligned gutters and file-level statistics showing insertions and deletions, so reviewing an agent's output feels closer to reviewing a pull request than reading a chat transcript. The sessions list adds compact quick-chat rows, making it easier to track several concurrent agents without losing the thread.
The more strategic addition is worktree support for Claude and Codex agents, which was previously limited to Copilot. Git worktrees let each agent operate on an isolated checkout of the repository, so parallel agents do not step on each other's changes. Extending that across harnesses signals that Microsoft is building the Agents window as a neutral cockpit for whatever model a team prefers. For organizations running a mix of coding agents, having one review surface that treats them consistently reduces the tooling sprawl that multi-agent adoption tends to create.
TypeScript 7 Compiles the Editor Itself
Quietly, VS Code 1.130 ships a proof point for the TypeScript rewrite. Microsoft compiled the VS Code repository using the TypeScript 7 release version, and the built-in extension was also updated to TypeScript 7. Using the new native compiler to build one of the largest and most scrutinized TypeScript codebases in existence is a strong statement that the rewrite is ready for production workloads, not just benchmarks.
The practical payoff for teams is compile and type-check speed on large projects, where the previous compiler had become a genuine bottleneck. Engineering organizations that maintain sprawling TypeScript monorepos have a concrete reason to start evaluating TypeScript 7 against their own build times. When the editor team eats its own dog food at this scale, it removes a common excuse for waiting. The migration risk is now easier to justify to a skeptical staff engineer who wants evidence before touching the build.
Spend Visibility Arrives Inside the Editor
VS Code 1.130 adds an aggregate AI credit usage display for Copilot Business and Enterprise customers, surfacing running consumption directly in the editor. Combined with new chat timestamps that track elapsed time per turn, the release gives individual developers and their managers a clearer picture of what agent activity actually costs and how long tasks take. This is the second consecutive Microsoft release cycle to foreground agent spend controls.
The signal for platform and engineering leaders is that agent cost is becoming a first-class operational metric rather than a line item that shows up on an invoice weeks later. As teams shift more work to autonomous coding, per-developer credit visibility helps managers spot runaway usage before it becomes a budget conversation. Pairing spend visibility with the elapsed-time data also lets teams reason about the cost per completed task, which is the number that actually matters when justifying continued investment in agent tooling.
Small Quality Wins That Add Up
The release also cleans up a long-standing terminal annoyance. File links in git diffs are now clickable when git's diff.mnemonicPrefix option is enabled. VS Code recognizes prefixes like i/ for the index and w/ for the working tree, strips the prefix from the target, and opens the correct file. It is the kind of fix that saves a few seconds hundreds of times a week for anyone who lives in the integrated terminal.
Taken together, VS Code 1.130 reads as an infrastructure release dressed up as a point update. The Agent Host Protocol, cross-harness worktree support, and assisted approvals are all foundations for running many agents reliably rather than demonstrating one. For engineering organizations deciding how deeply to standardize on VS Code as an agent platform, this cycle is the one that makes the editor look like a durable place to build that workflow rather than a temporary host for it.
