Intel / post

Claude Managed Agents Adds a Session Viewer to the ant CLI

Source: x.com
Date: Published Sep 10, 2026

From the Daily Brief

Anthropic added two features to Claude Managed Agents: a session viewer reached with ant beta:sessions connect, including a web interface, and an auto mode that reviews each tool call against the agent's stated intent to decide whether to run it, deny it or ask for approval. Together they target the two things that make long-running agents hard to trust in production: visibility into what a session is actually doing right now, and a way to automate approvals without removing oversight entirely. Before this, monitoring a long-running Claude agent meant tailing logs or interrupting the session to check its state, and automating approvals meant either allowing everything or reviewing every call by hand. The auto mode sits between those extremes by judging each call against stated intent rather than a fixed allowlist, which should cut manual review load on routine calls while still catching ones that drift from the task.

Read the 2026-09-11 Brief →

Context

Anthropic's `ant` command-line tool can now attach to a Claude Managed Agents session that is already running, the ClaudeDevs account announced. The command is `ant beta:sessions connect`, and adding `--web` opens a browser interface served from the developer's own machine.

The viewer is for steering as well as watching. Anthropic's documentation says the terminal view loads the session's transcript and follows it live, and the developer can send a message, interrupt the , or allow or deny a tool call that is waiting for approval. Detaching leaves the session running, and reconnecting loads its full history. The browser version offers the same controls and, unlike the terminal view, follows every thread when the agent delegates work to other agents. According to the docs, that local page sends requests only to the running `ant` process, so credentials stay in the CLI, and the URL it prints can be opened once, within two minutes.

The command extends the same CLI that earlier gained `ant apply`, which defines Managed Agents resources as files. The post introduces two updates, and only this first one is covered here; the attached demo video was not reviewed.

Terms in this piece · Glossary

AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.

Read the source x.com