Hacker News Tracks Claude Code's Web Scheduler as Anthropic Pushes Recurring Agent Work Into the Cloud

What Anthropic launched

On March 27, 2026, Anthropic documented cloud scheduled tasks for Claude Code on the web. The feature lets a user run a prompt on a recurring cadence using Anthropic-managed infrastructure, so the job keeps running even when the user's own machine is off. The documentation positions it for recurring development chores such as morning pull request review, nightly CI failure analysis, documentation sync after merges, and weekly dependency audits. That framing matters because it moves Claude Code from an interactive assistant into something closer to a background operator.

Anthropic says the feature is available to Claude Code on the web users across Pro, Max, Team, and Enterprise plans. A task can be created from the web interface, from the desktop scheduler, or from the CLI via /schedule. The key operational detail is that every cloud run starts in a fresh session with a fresh clone of the selected GitHub repositories. Claude begins from each repo's default branch and, by default, pushes only to claude/-prefixed branches unless the user explicitly loosens that policy.

How cloud scheduling differs from local automation

The docs compare three scheduling modes: cloud tasks, desktop tasks, and session-scoped /loop. Cloud tasks do not need the user's computer to stay on, survive restarts, and do not require an already open session. In exchange, they do not get direct access to local files. Desktop tasks and /loop keep local access, but they depend on the user's own machine and, in the case of /loop, on the active session itself.

Anthropic also exposes a fairly explicit control plane around the feature. Each task chooses a cloud environment that can define network access, environment variables, and a setup script. MCP connectors such as Slack, Linear, or Google Drive are included by default and can be removed per task. The minimum interval is one hour in the web scheduler, while custom schedules can be refined from the CLI. Each run creates a new visible session where the user can inspect the work, review code changes, and turn the result into a pull request.

Why the HN thread mattered

The Hacker News submission reached 282 points and 230 comments at crawl time, which suggests the release landed as more than a minor convenience feature. The discussion naturally centered on trust boundaries: what happens when agentic work becomes recurrent, autonomous, and connected to repositories and external services. Anthropic's design choices, especially fresh clones, scoped branch pushes, environment-level controls, and replayable sessions, read like attempts to make recurring agent execution operationally acceptable rather than merely possible.

That is the larger significance. Claude Code already had local automation options, but cloud scheduled tasks make recurring agent work first-class and durable. If the workflow proves reliable, engineering teams may start treating code assistants less as tools they open and more as background systems they supervise.

Primary source: Claude Code Docs. Community discussion: Hacker News.