Claude Projects from your phone still isn't a shop line

Anthropic redesigned Claude Projects on September 17, 2026. You describe a goal, a coordinator splits the work into parallel threads, and you can steer it from your phone after you leave the laptop. That is a real product change. It is still a coding desk, not a shop line.

I sell owned deployments. The useful question is what this removes from your stack, and what it does not.

What did Anthropic change in Claude Projects?#

Old Projects were a folder. New Projects, in beta inside Claude Code, are a coordinator plus worker threads that keep going after you close the computer. You pick a goal and a repo or a document set. Claude scopes the job, opens threads, reviews the output, and assembles the result. You steer from the main project chat, or you drop into one thread.

Anthropic’s own examples are checkout latency across endpoints and retiring a deprecated v1 API across web, API, and mobile repos. Threads run as separate Claude Code cloud sessions on their own branches. Overlap shows up as a merge conflict, the same way two humans colliding on a PR would. Documents get read and drafted. Repos get tests and pull requests.

Access is narrow on day one: select Claude Pro and Max subscribers who already use Claude Code cloud sessions and have no existing projects on web or desktop. More Pro and Max Claude Code users over the following week. Chat, Cowork, Team, and Enterprise come later. Existing projects keep working until that upgrade lands. Threads run in the cloud today; local tools behind your network are “very soon,” with no date.

Shared memory is the other change. A thread can remember the release moved to Friday or who to ping before touching billing, and a library stores files you added plus artifacts Claude produced. Useful context. Not a CRM.

Does steering Claude Projects from a phone replace an owned agent?#

No. Checking a coordinator chat on a phone is not the same as an owned console that writes CRM notes, texts a lead back, or sits on the business number. The redesign is for people shipping code. If your “project” is a salon book, a Jobber board, or a law-firm intake queue, this beta does not touch it.

Use it if you already pay for Pro or Max, you live in Claude Code, and the pain was juggling sessions by hand. Do not buy a custom agent to open pull requests. Do buy a custom agent when the work has to live in a chat your staff already uses, with your keys, a kill switch, and no Anthropic UI in the middle.

That is the same line I drew after the Claude Cowork and chat merge: a rented desk that got better. A Telegram AI Agent is a $2,000–$4,000 one-time owner console you keep. If SuperGrok Heavy or Cursor Ultra already covers scheduled computer-use work, compare against that bundled plan first.

If you are mapping AI for small business and the leak is missed calls, skip this announcement. The phone still rings whether or not a cloud thread is profiling an endpoint.

What still sits outside this redesign?#

It does not answer the shop line, it does not write to GlossGenius or Housecall Pro unless you already wired that yourself, and several full Claude Code sessions will chew a usage cap faster than a single chat. Anthropic is explicit about that last point. You can pick models and effort levels for the coordinator and the workers. That is a cost control, not a receptionist.

Do not treat project memory as the system of record. Jobber, HubSpot, a shared sheet, or the appointment book still owns the truth. If a customer is on the line, this product is silent.

This post is going out from the Discord agent I sell. Claude Projects from phone is a good Claude Code day. It is not a reason to confuse a parallel coding floor with an owned operator console.

If you want the boundary mapped — keep Claude Projects for the repo, own Telegram for the phone-side console, or leave both alone — send the short audit form. I reply with your AI replacement map within 24 hours.