Claude Cowork launches on mobile and web for Max users

• Claude Cowork is moving onto mobile and web in a beta that starts with Max, with more plans later, according to Claude's rollout note.
• Background work is now explicit product behavior: scheduled tasks run with the computer off, and judgment calls can be pushed to the phone, per Claude's scheduled-tasks note.
• Chat and Cowork are merging into one home on web and desktop, with shared Projects and Artifacts, as Claude's navigation update says.
• The engineer-facing plumbing is already landing in Claude Code 2.1.202: Remote Control fixes, workflow-size config, and OpenTelemetry attributes appear in Claude Code Log's changelog.
• Doubled Cowork usage limits run through Aug. 5, per Claude's limit update.

The mobile UI in mikeyk's post looks less like chat and more like an inbox: Needs input, Ready for review, Approve once, Respond to question. TestingCatalog's Ask Mode screenshot exposes the permission split: Ask before acting or Act without asking. _catwu's sourcing workflow is the cleanest demo of the product shape: kick off a workflow, generate an artifact, email it, lock the laptop, review on the go.

Claude said Cowork is rolling out to mobile and web over the next several weeks, starting with Max, with more plans to follow. The handoff is the product: assign work at a desk, close the laptop, pick up the result from a phone.

TestingCatalog said the first mobile and web rollout was reaching Max subscribers and that the new surfaces can trigger, follow up on, and schedule Cowork tasks in its rollout note. Wes Roth's summary put the same bundle together: sessions and files move across devices, tasks keep running after the laptop closes, and questions can arrive on the user's phone in his launch summary.

Scheduled tasks now run even when the computer is off. Claude's example is a 6 a.m. client-prep task that works through threads and transcripts, builds a brief, and leaves a follow-up drafted but unsent.

The important verb is not "chat." It is "leave": Claude leaves a draft, leaves a question, leaves a task state that can be reviewed later.

rohanpaul_ai said Anthropic paired the rollout with usage data from 1.2 million anonymized Cowork sessions across more than 600,000 organizations between May 11 and May 31, and that most Cowork activity sat outside coding in his analysis. He described the target work as the messy layer spread across email, docs, notes, calls, spreadsheets, and reminders.

Chat and Cowork now share a home tab on web and desktop. Claude described it as one place for Projects and Artifacts across both modes in its navigation update, while mikeyk added the concrete UI pieces: one sidebar, one search, and one place for Projects and Artifacts in the desktop/mobile screenshot post.

The split between "I'm building something" and "I'm working on something" is getting thinner. Mat Velloso's reaction framed Claude's single-app approach against OpenAI's split between ChatGPT and Codex: writing code, editing documents, and using tools are becoming the same surface with different risk levels in his comparison.

TestingCatalog found an Ask Mode selector in the mobile Cowork UI in its mobile UI post. The menu has two modes:

• Ask before acting: "You'll confirm before anything changes."
• Act without asking: "Claude works without pausing for approval."

That matches Claude's official phone-handoff language: when a decision needs judgment, the question goes to the phone in the scheduled-tasks note. The mobile screenshot in mikeyk's post shows the same approval vocabulary, with tasks marked Approve once and Respond to question in the Cowork inbox image.

Claude extended doubled Cowork usage limits through Aug. 5 and pointed users to the Cowork announcement. TestingCatalog said the mobile and web rollout starts with Max subscribers and that other plans will get it within the coming weeks in its rollout note.

rohanpaul_ai added one quota detail for operators: Cowork usage counts against the same shared usage pool across Claude.ai, Claude Desktop, Claude Code, and related surfaces in his analysis. That makes Cowork another consumer of the same plan budget rather than a separate agent quota.

Claude Code 2.1.202 shipped the unglamorous pieces that make mobile and web control less brittle. The changelog lists 18 CLI changes and one system-prompt change.

The Remote Control fixes are the ones that line up with Cowork's new surfaces:

• Commands sent from mobile or web into an interactive session no longer fail with "Unknown command" in the full changelog.
• Images and files sent from Remote Control without a caption are no longer silently dropped in the full changelog.
• Remote Control sessions now show the correct permission mode in mobile and web apps in the full changelog.
• Background session renames no longer revert when a job restarts in the full changelog.

The same release added a Dynamic workflow size setting in /config, with small, medium, and large agent counts as an advisory guideline rather than a hard cap in the full changelog. It also added two OpenTelemetry attributes for workflow-spawned agents so a workflow run's activity can be reconstructed from telemetry Claude Code Log's summary.

HamelHusain surfaced the always-on Remote Control flag in ~/.claude/settings.json in his config note:

He also called it brittle, citing a hang when changing /effort on the host while using the client in the same note.

_catwu's candidate-sourcing workflow shows how Cowork, Claude Code workflows, artifacts, and mobile review fit together:

1. Describe the role and target backgrounds.
2. Kick off a dynamic workflow to find 100 candidates.
3. Include LinkedIn, Twitter, blogs, podcasts, and a one-line pitch for each person.
4. Ask Claude Code to make an artifact and email it.
5. Lock the laptop and review the finished list on the go.

In replies, _catwu said the workflow used computer use in a follow-up and took about 20 minutes with workflows in another reply. The artifact mattered enough that _catwu later said it was faster to read than Markdown in a reply about artifacts.

Corey Ganim described the onboarding flow he uses with AI Concierge clients. It reads like a field manual for turning Cowork from a blank agent into an office-worker harness:

1. Connect tools first: Google Drive, Gmail, Calendar, Slack, Notion.
2. Create three context files: about-me.md, brand-voice.md, working-style.md.
3. Set global instructions that apply to every session.
4. Introduce plugins or skills, with Productivity and Marketing as examples.
5. Demo scheduled tasks, often with a morning brief.

The order is the technical point. Ganim connects tools before writing context files because Cowork can pull from those tools while creating the context in his setup process.

TestingCatalog said recent Claude iOS builds contain new Conway icons and a "Waking Conway..." screen. It described Conway as Anthropic's expected always-on AI agent running in a separate cloud container in the Conway post.

The same post mapped the four-tile icon to apps or custom UI tabs, possibly a new extension standard, and mapped the scroll icon to Memory with "Dreams," a feature TestingCatalog said was released for Claude Managed Agents in the Conway post. Conway is separate from the Cowork rollout, but it points at the same product grammar: a persistent cloud agent, tool tabs, memory, and phone-first control.