For most Australian firms, nothing changes today — but the shift underneath this update does matter. Anthropic has rebuilt Claude Projects so one conversation can coordinate several pieces of AI-driven work at once, checking in only when it needs you. That is the same pattern AI Smarter already builds into AIOS for professional-services firms: describe the outcome once, let AI run supervised threads, review the result.
Four things are worth knowing before the detail below:
Anthropic announced the redesign on 17 September 2026, on X and in an updated product page.
It is in beta for Pro and Max subscribers who already use Claude Code cloud sessions.
It is not yet available on Team or Enterprise plans.
Each "thread" of work runs as its own Claude Code cloud session, tied to a connected code repository.
What did Anthropic actually announce?
Claude Projects has moved from a filing system to a coordinator. A "project" used to be mostly a place to group related conversations and files — you still had to run, track and stitch the work together yourself.
Anthropic's own product documentation now describes a project as "one ongoing conversation where Claude coordinates a stream of related work for you." You tell it what needs doing, and "it starts a thread for each task." Each thread is a Claude Code cloud session — Claude Code (Anthropic's coding-and-automation agent) running on Anthropic's servers instead of your machine. "Threads run in parallel and keep going after you close the laptop, and you can check on them and steer them from your phone," the documentation states.
The company's own account, @claudeai, summarised it more plainly on X: "Projects now run from one conversation, starting in Claude Code. You describe what needs doing, and Claude directs parallel threads that keep working after you close your laptop."
How is this different to just using Claude normally?
Without a project, running two pieces of work in Claude at once means running two separate conversations. You do the coordinating yourself: deciding what each one covers, repeating the same background twice, and checking back on each one.
A project removes that manual step. Per the documentation, you "send work to one place": paste in a task whenever one comes up, and Claude either starts a new thread for it or hands it to a thread already working in that area. It answers quick questions in place.
The practical shift is that Claude becomes the coordinator, not just the assistant answering one question at a time. It scopes each request, delegates it, reviews what comes back, and checks in with a person only when a real decision is needed.
Access is limited and rolling out gradually
The documentation confirms Projects are "in public beta on Pro and Max plans and rolling out gradually, starting with accounts that have used cloud sessions and don't have existing projects in claude.ai chat or Cowork," and that they "aren't available on Team or Enterprise plans yet."
For most Australian SMB owners reading this, the honest answer is: not yet, and not automatically. This is a developer-facing feature, built around Claude Code and cloud sessions connected to a GitHub repository. It isn't a change to the general chat-style Claude most business owners use day to day.
What would this actually look like inside a professional-services firm?
Anthropic's own framing is technical — coordinating a change across multiple codebases from one instruction. That's a developer's problem, not a law firm's or an accounting practice's.
But the pattern underneath it fits professional services well: describe the outcome once, let an AI system split the work into parallel threads, review what each produces, and interrupt a human only for a genuine decision. That is exactly the model AI Smarter builds through AIOS, our AI Operating System, for firms with no developers on staff.
Take a conveyancing practice. One thread could chase outstanding disclosure documents from a vendor's solicitor. A second drafts a client status update. A third checks a settlement date against the calendar. All three report into one daily briefing, instead of three inboxes someone has to babysit. An accounting practice already runs a version of this without calling it "threads" when it onboards a new client: one person requests identity documents, another sets up the ledger, a third books the first meeting. This announcement signals that mainstream AI tools are moving toward automating that coordination role itself, not just the tasks sitting inside it.
The bigger risk here is judgement, not capability
Giving a system permission to keep working unattended raises the same question any delegation does: what can it do without asking first, and what must wait for a person? Claude is designed to check in "when it needs you" and let an owner "steer any thread, even from your phone." But where that line sits — between "needs a decision" and "should have asked first" — is set by how the system is configured, not by the technology alone.
There's a second, more practical limit. This version of Projects is built around cloud sessions tied to a connected code repository. A firm's templates, client checklists or house rules that only live on a local machine won't automatically carry into a cloud thread. That's the same lesson we give every client before switching on any automation: know precisely which systems and documents an AI tool can see, and which it can't, before handing it a multi-step task with real client consequences.
Does a tool like this replace a structured AI system for my business?
No — if anything, it makes the case for having one. A firm running AIOS already has a single system watching CRM (customer relationship management software), email, accounting, documents and marketing. It coordinates what happens across them and briefs the owner once a day, rather than leaving them to check ten separate dashboards.
Anthropic's redesign is evidence of where the whole industry is heading: fewer standalone conversations, more coordinated, reviewed, delegated work. What changes for a business owner isn't whether that coordination is worth having. It's which system is disciplined enough to do it on their behalf, with guardrails suited to a regulated professional-services practice.
What should you actually do about this now?
Nothing urgent — but it's worth watching rather than ignoring. If your firm doesn't use Claude Code or store its work in a connected code repository, this specific feature won't reach you for some time, and may never apply in its current, developer-oriented form.
What matters is the pattern underneath it: AI shifting from answering one question to running a supervised piece of work end to end, checking in only when it must. That's the shift AI Smarter has spent the past year building AIOS around for Australian professional-services firms, well before this particular announcement. We cover these translations regularly on the blog as they happen, precisely because most of them arrive dressed as developer news first.
If you want a plain-language read on what a properly supervised, coordinated AI system would look like inside your own practice — with the review points and guardrails a regulated business needs — book a free initial AI consultation and we'll walk through it against your actual tools, not a generic checklist.