Multi-Chat in VS Code 1.128 — Running Claude Agents in Parallel

VS Code 1.128 has landed

Visual Studio Code 1.128 shipped on July 8, 2026. One of its headline features is multi-chat support inside Claude agent sessions. A single session can now hold several related chats, letting you compare approaches, branch from an earlier turn, and run work in parallel.

It looks like a modest UI change, but in real AI-assisted coding the ability to handle branching context has a large effect on efficiency. This post looks at what the feature actually solves, from Webhani's perspective.

The problem multi-chat solves

Until now, AI coding meant piling context into a single conversation. That is convenient, but it gets awkward in a few common situations:

• You want to compare implementation A and implementation B, but trying both in the same conversation muddies the context.
• You want to rewind to an earlier point and try a different direction, while keeping the reasoning that came before.
• You want to write the implementation and its tests at the same time.

Multi-chat in VS Code 1.128 addresses these by hanging several chats off one session. Each chat keeps its own history, title, and model selection, and restores with the parent session after a restart. Peer chats stay grouped under the session and do not clutter the top-level session list.

Three interaction models

Based on the official notes, there are three main operations.

1. Adding a peer chat

You can add parallel chats to the same session. For example, the main chat adds a /health endpoint to an Express app while a peer chat writes tests for it at the same time.

Claude session: improve user-service
├─ Main chat   : add the /health endpoint
├─ Peer chat #1: write tests for it, in parallel
└─ Peer chat #2: explore an alternative implementation

2. Forking

You can branch from a turn partway through a conversation. In 1.128, a fork creates a new peer chat within the same session rather than a brand-new top-level session. The forked chat inherits the conversation up to the fork point, runs independently from its siblings, and gets an auto-generated title.

This fits the "I want to share the understanding up to here, but go a different way from this point" situation.

3. Concurrent turns

You can send turns to multiple chats at once. Instead of waiting on one reply at a time, you can advance several pieces of work in parallel, which cuts idle time.

Where it earns its keep

Here are the situations where we expect this to help when using AI coding on client projects.

• Separating implementation from verification. Keeping the implementation chat apart from a test-and-review chat avoids cross-contaminating context.
• Comparing designs. Running two approaches to the same requirement in separate chats and comparing the output becomes a natural way to decide.
• Investigation alongside implementation. Keeping a chat for probing a library's behavior separate from the implementation chat means a long investigation thread does not pollute the implementation context.

There is a caveat, though. The more you parallelize, the more output there is to review. Whether a human can actually verify every line the AI generates in parallel is a separate question. At Webhani we recommend designing "build fast in parallel" and "pass it through a verification gate" as a pair. Parallelism raises throughput; it is not a reason to loosen your quality gates.

Operational considerations

If you use this on a team, deciding a few things up front reduces confusion.

1. Chat naming. Auto-generated titles are handy, but when you later trace back through history, a name that tells the team what a chat was for makes review easier.
2. Model selection per chat. Since each chat can pick its own model, you can route cheap, fast models to lightweight investigation and stronger models to hard implementation — building the cost/quality trade-off into your workflow.
3. Clear ownership of verification. With more work running in parallel, do not leave the scope of human review ambiguous. The more output you generate, the more this boundary matters.

Takeaways

Multi-chat in VS Code 1.128 widens AI coding from "a single conversation" into "a workspace you can branch and parallelize." With forks now becoming peer chats inside the same session, it is easier to split directions while preserving context.

That said, parallelism alone does not guarantee results. The faster you can build, the more your verification design matters. At Webhani, we track these tool developments and help clients design workflows that build in parallel and verify reliably.

References: Visual Studio Code 1.128 (code.visualstudio.com), Claude AI Gets Yet Another Boost in VS Code 1.128 (Visual Studio Magazine)