GLM-52 897 —
GPT-56SC 873 —
CL-OP5X 865 -0.9%
GROK-46H 865 -0.9%
GEM-37FH 865 -0.9%
GPT-56T 861 —
GLM-5 856 —
MUSE-SPK 841 —
QWEN-38X 824 -2.3%
GPT-6A 820 —
KIMI-K3X 810 -1%
CL-FAB5H 787 -0.9%
CL-OP5H 764 -0.9%
CL-OP46H 742 -0.9%
CL-OP47H 733 -1.1%
GEM-38FH 676 -1%
CL-OP47 585 -0.7%
INKL 531 —
CL-OP46 496 -0.2%
CL-OP48 490 -0.2%
GLM-52 897 —
GPT-56SC 873 —
CL-OP5X 865 -0.9%
GROK-46H 865 -0.9%
GEM-37FH 865 -0.9%
GPT-56T 861 —
GLM-5 856 —
MUSE-SPK 841 —
QWEN-38X 824 -2.3%
GPT-6A 820 —
KIMI-K3X 810 -1%
CL-FAB5H 787 -0.9%
CL-OP5H 764 -0.9%
CL-OP46H 742 -0.9%
CL-OP47H 733 -1.1%
GEM-38FH 676 -1%
CL-OP47 585 -0.7%
INKL 531 —
CL-OP46 496 -0.2%
CL-OP48 490 -0.2%
← Back to feed

Claude Cowork Goes to Web and Mobile — and 90% of Its Sessions Have Nothing to Do With Code

Claude Cowork launched in January as a desktop application. As of this week it is available on web and mobile for Max subscribers — a platform expansion that coincides with a data disclosure that reframes what the product actually is.

Anthropic reports that 90% of Cowork sessions are not coding work. Users are running research, drafting documents, planning projects, and doing analysis. The Claude Code lineage — the underlying agentic scaffolding — is doing general knowledge work at scale.

What Boris Cherny Said at the Developer Conference

At Anthropic’s annual developer conference in San Francisco, head of Claude Code Boris Cherny described the productivity shift his team has experienced directly:

“About six months ago, all the code I used to write by hand, now Claude writes. I prompt Claude, it builds the feature, tests it, shows me. I say ‘yes’ or ‘make this change’ and it makes the change.”

The headline number from the same session: the volume of code written per engineer at Anthropic has grown approximately 250% since the introduction of Claude Code, with code quality and reliability holding stable.

Cherny’s team identified five emerging archetypes for how engineers work alongside AI agents — and noted that the largest productivity gains may not come from senior engineers. Accountants, marketers, and new graduates who can direct agents toward clear goals may benefit more than veteran developers whose existing workflows are already optimised.

Beyond the Coding Tool Frame

The 90% non-coding figure suggests something structurally different from a dev tool. Cowork’s underlying model — Claude running with persistent context, file access, and extended autonomy — turns out to be well-suited to the research-and-synthesis workflows that make up most knowledge work. The desktop-first launch in January created a ceiling on this: agents that need to read documents, open web pages, and iterate over hours are poorly served by a phone-only or browser-tab experience.

The web and mobile launch removes that ceiling. Max subscribers can now initiate Cowork sessions from any device, hand off longer tasks to run while they’re away from their computers, and resume on mobile.

Pricing Context

Claude Cowork is available to Claude Max subscribers. Max starts at $100/month. The platform expansion does not change pricing.

Claude Code — the developer-facing terminal agent — remains separate. The two products share underlying model infrastructure but target different primary workflows.

The separation matters: Cowork’s 90% non-coding usage rate suggests Anthropic is building the general-purpose agentic platform it originally positioned Claude Code as, while Claude Code itself narrows toward professional software development with deeper IDE integration, SWE-Bench optimisation, and enterprise deployment controls.