← Back to all drops

Drop #29: The Cloud Switch Lands

by ClaudeCowork.com

Last week we covered what was coming on October 6. This week it arrived: new Cowork tasks on Pro and Max now run in the cloud. The desktop app shipped three releases around the switch, with changes worth knowing if you schedule tasks or let Claude use your computer. Elsewhere, Claude Code got mods, Claude for Government reached general availability, Anthropic expanded Claude for Startups, and it committed $100 million to training engineers who deploy Claude inside companies.

October 6 Came and Went: Cowork Runs in the Cloud

Anthropic's support article on using Cowork across web, desktop, and mobile set October 6, 2026 as the date new Cowork tasks on Pro and Max plans start running in the cloud and the Only on your computer option leaves Settings. Tasks you started on your computer before the switch stay there until you finish them. The notice is addressed to Pro and Max plans.

The October 6 desktop release (2.26454.0) cleaned up one rough edge of the move: a scheduled task that had been moved to the cloud could show up as a separate paused task on your computer when cloud tasks failed to load, and turning that copy on could run the task twice. That is fixed. The same release fixes the app crashing shortly after launch for people with a very large number of Cowork tasks, and the app freezing when a second message was sent to a new task before it had finished starting.

Scheduled Tasks Claude Creates Now Auto-Approve by Default

One line in the October 1 desktop release (2.19675.0) changes how scheduled work behaves: scheduled tasks that Claude creates for you now use Automatically approve by default where your organization allows it. Their runs use tools without asking first and pause only when something looks unsafe. With tasks now running while your laptop is closed, that is what makes an unattended schedule actually finish - and it is also a reason to look at what each scheduled task is allowed to touch.

Computer Use: Safer Defaults and Better Background Clicking

Three changes to computer use in Cowork landed this week. On October 1, computer use on macOS 13 and earlier was removed, because its screenshots could never be taken there; it now requires macOS 14 or later, and the Computer use section in Settings explains why when it is unavailable. The same release fixed Claude clicking the wrong item, missing an open menu, or bringing the app to the front in apps such as Slack while working in the background.

On October 6, computer use on macOS stopped editing or saving hidden files and folders in your home folder, such as ~/.vimrc or ~/.aws. That release also fixed Claude being unable to use document menu commands, such as New Slide or Undo in Keynote, or to click and drag on some canvases while the app stays in the background.

Claude Code Gets Mods

On October 1, Anthropic introduced mods: small TypeScript functions that change how Claude Code works. A mod hooks into an event - a tool call, a permission request, part of the screen being drawn - and can run before it, after it, or instead of it. That lets a mod rewrite a prompt, block or retry a tool call, redact secrets from tool output, or add buttons and panes to the interface. Mods work in the Claude Code CLI and the desktop app, and you can ask Claude Code to write one for you.

Mods ship inside plugins, so they install and share like any plugin and fall under the same admin controls. Anthropic is clear that mods are not sandboxed and run with the same access to your machine as Claude Code, so install them only from sources you trust. On Team and Enterprise plans, and anywhere managed settings are deployed, a built-in mod called sec-default loads first and stops user-installed mods from doing risky things such as overriding permission deny rules. Some built-in features are now mods themselves - the /diff feature is one - so you can turn them off or replace them.

Claude for Government Is Generally Available

On September 30, Claude for Government moved from public beta, where it had been since July, to general availability for federal and state agencies. It runs in a FedRAMP High authorized environment, and Claude works directly with files on the desktop so agency staff can use skills, plugins, and projects for memos, RFP reviews, and casework. There are no seat fees: agencies buy usage in fixed increments with a hard not-to-exceed cap. The Claude Code CLI and Claude for Microsoft 365 are rolling out in early access through the same environment.

A Free Year of Claude Team for Startups, and $100M for Training

On October 6, Anthropic expanded Claude for Startups, as reported by TechCrunch. Qualifying companies get a free year of Claude Team with up to five premium seats, $1,000 in API credits, access to Claude Marketplace, and virtual office hours with Anthropic's Applied AI team. Companies are eligible if they were founded in the last five years or received funding in the last two; applications go through the Claude for Startups program page.

On October 2, Anthropic launched Claude Frontier Academy, backed by a $100 million commitment to train 10,000 Frontier Deployed Engineers by the end of 2027. Its first program is a residency: a multi-day in-person course, then 12 weeks leading a real Claude project at the engineer's own organization. Participation is by nomination, and the first cohorts include engineers from Accenture, Bain, Capgemini, Commonwealth Bank of Australia, Deloitte, McKinsey, Morgan Stanley, and Novo Nordisk.

Why This Matters for Cowork Users

The cloud switch was the headline, but the quieter changes decide how it feels day to day. Scheduled tasks that run with nobody watching now approve their own tool use by default, and computer use got stricter about what it will touch on your Mac. Together they point to the same habit: decide up front what a task may do, because you will not be there when it runs.

Tip of the Week: Audit Your Scheduled Tasks

Scheduled tasks that Claude creates for you now use Automatically approve by default where your organization allows it, and their runs pause only when something looks unsafe. That is convenient, but worth a once-over.

❌ Weak

Every Monday, clean up my inbox and reply to anything urgent.

✅ Strong

Every Monday at 8am, read my unread email from the last 7 days and save a summary of anything that needs a reply to a new doc. Draft replies, but do not send anything, delete anything, or change any settings. If a message looks like it needs a payment or a password, list it and stop.

The rule of thumb: a task that runs unattended should say what it may change and what it must leave for you. Spell out the limits in the task itself, then check the task's approval setting matches how much you trust it.

Worth Reading This Week

Until next time.

— The ClaudeCowork.com team