Claude Hub has quietly slipped from an internal experiment into Early Access, according to app teardowns shared on 1 September 2026 — and in the same breath, a placeholder tile for Managed Projects has appeared in the layout code of Claude’s mobile app. Neither feature has been announced by Anthropic. Both have now been tracked across six weeks of sightings, and together they sketch the clearest picture yet of where the company is taking Claude: away from a one-to-one chat window and towards an orchestration surface where sub-agents work under your supervision.
This article follows up our earlier report on the first Hub Mode sighting, when the feature was still flagged as internal. Since then the flag has changed state, the feedback prompts have started, and a second feature — Managed Projects — has edged towards mobile. Everything below is sourced from published leaks and official Anthropic documentation, and we flag clearly which is which. If you are planning how agentic AI fits your business, the Claude Hub story is worth ten minutes of your attention.
Table of contents
- What Is Claude Hub?
- The Road From Hub Mode to Claude Hub Early Access
- Managed Projects: Claude’s Autonomous Workspace
- What the Mobile Sighting Really Shows
- What Claude Hub Could Change Day to Day
- How Claude Hub Fits Anthropic’s Product Arc
- Claude Hub vs ChatGPT Projects vs Gemini AI Rooms
- Who Gets Claude Hub First? Plans and Pricing Signals
- What Your Business Should Do Before Claude Hub Ships
- What Is Still Unconfirmed
- References
What Is Claude Hub?
Claude Hub is Anthropic’s name for an unreleased control surface that orchestrates multiple Claude sub-agents from one place. The evidence comes from TestingCatalog, the outlet that specialises in unreleased app features, which reported the Early Access transition on 1 September 2026.
A control surface for sub-agents
The first sighting, on 18 August 2026, described a “Hub Mode” toggle inside the Claude mobile app, marked internal, sitting next to a planet icon. Its description said the mode “will apply to new tasks and requires users to restart the app after activation.” A related Hub destination inside Claude Code on the web showed a list of sub-agents; selecting one displayed a message that Claude “will generate a task board for the user and their agents.” In plain terms: one screen, many agents, and a board that tracks what each is doing.
From a planet icon to a task board
The planet icon has its own history. It previously carried the label Orbit, a briefing-tool concept reported in May 2026, before that label was removed. That lineage is a hint, not proof — but it suggests Claude Hub absorbed an earlier idea about giving users a standing overview of work in progress, rather than a scrolling transcript of one conversation.
What Early Access actually means here
The phrase “Early Access” in this story is a feature-flag stage, not a programme you can sign up for. On 29 August the flag was still internal, with a prompt asking testers for feedback if they switched back to a normal chat. By 1 September the flag read Early Access. Anthropic sells early access as a benefit of its Max plans, which start at $100 per month and explicitly advertise “early access to advanced Claude features” — but which cohort gets Claude Hub first is not confirmed anywhere.
The Road From Hub Mode to Claude Hub Early Access
Six weeks of sightings turn a single screenshot into a trajectory. The table below reconstructs the timeline from the published record, so you can judge the pace yourself.
| Date | Sighting | Status |
|---|---|---|
| 22 Jul 2026 | Managed Projects screen spotted: standard vs “managed” project choice | Internal, staff and trusted testers |
| 16 Aug 2026 | Collaborative Claude Code Projects: repositories, threads, shared sessions | Internal development |
| 18 Aug 2026 | Hub Mode toggle found in the Claude mobile app, planet icon, restart required | Internal flag, not operational |
| 29 Aug 2026 | Hub testing continues; feedback prompt on switching back to normal chat | Internal flag |
| 1 Sep 2026 | Hub flag transitions to Early Access; Managed Projects tile appears in mobile layout | Early Access flag, unannounced |
Why the pace matters
Feature flags can sit unchanged for months. This one moved from internal to Early Access in fourteen days of visible testing, with feedback collection in between — the pattern of a feature being actively prepared for a cohort, not a shelved prototype. That is why the Claude Hub transition is the most newsworthy line in the table, even though it is the least visual: flag stages are how modern apps stage a launch.
Why the Orbit lineage matters
Renamed features usually keep their plumbing. If Claude Hub inherited Orbit’s briefing concept, the task board is likely to summarise agent activity for you, rather than expecting you to poll each sub-agent. For anyone who has managed a team of human contractors, the analogy is direct: a stand-up report, generated continuously, for a workforce that happens to be software.
Managed Projects: Claude's Autonomous Workspace
Managed Projects is the second thread of this story, and it is older than the Hub sightings. First spotted on 22 July 2026, it offers a choice at project creation between a standard project and a managed one, with a one-line pitch: “Claude takes on tasks and keeps the project organized.”
Shared memory and cloud sessions
According to the July teardown, sessions inside a managed project share memory and instructions, so context carries forward and Claude runs more autonomously. Managed projects only create and manage cloud sessions — the work happens in an isolated cloud environment where Claude can organise files, execute tasks and manage multi-step workflows. One analyst framing put it neatly: this gives Claude its own workspace, not just a chat window.
A Slack-like collaboration layer
An August sighting extended the concept: repositories as persistent context, multiple people sharing one session with Claude, and multiple threads as sub-conversations inside a project. TestingCatalog’s headline called it Anthropic developing “its own Slack.” Claude also periodically reorganises the project’s context files and memories itself — housekeeping as a background behaviour, which is exactly the kind of chore that makes shared workspaces decay when humans are responsible for it.
Who it is aimed at
The July report said Managed Projects is planned for Team and Enterprise plans, with personal and shared options. That matches the collaboration features in the August sighting: threads, shared sessions and repositories make little sense for a solo chat user, and a lot of sense for a working group that wants one agent-run home for a stream of work.
What the Mobile Sighting Really Shows
The 1 September post makes two claims, and the second is the weaker one: “Managed Projects for Claude Code are coming to Claude for mobile as well.” The evidence is an empty tile that now appears during layout loading in the Claude mobile app.
An empty tile is a hint, not a launch
An empty placeholder in layout code means engineers have reserved space for a feature — nothing more. It does not show working screens, it does not specify iOS or Android, and it can be removed in the next build. Treat it as a signal of intent with an unknown timeline. What makes it interesting is the pairing: a mobile tile for Managed Projects landing in the same window as the Claude Hub flag flipping to Early Access suggests the two are being staged together.
Why mobile matters for agent orchestration
Orchestration is a monitoring job, and monitoring is what phones are for. If Claude is running cloud sessions that continue while your laptop is shut, the natural companion is a mobile surface where you check the task board, approve a step, or redirect an agent from the sofa. A Claude Hub that only lived on desktop would undercut the whole premise of autonomous, long-running work. The mobile sighting, thin as it is, is the piece that makes the strategy coherent.
What Claude Hub Could Change Day to Day
Screenshots only go so far, so it is worth translating the leaked mechanics into an ordinary working week. Assume the pieces ship roughly as sighted: a Claude Hub task board, sub-agents, and managed projects running cloud sessions with shared memory.
A worked example: one brief, three agents
Picture a marketing manager opening a managed project called “Q4 launch” on Monday. Claude proposes a task board: one sub-agent drafting landing-page copy, one assembling a competitor pricing summary, one turning last quarter’s call notes into objection-handling points. Each runs in its own cloud session, all three share the project’s memory and instructions, and the Claude Hub board shows their state. The manager’s job shifts from typing prompts to reviewing outputs and unblocking stuck tasks — the same shift a team lead makes when they stop doing every task themselves.
The failure modes to expect
The honest version of that picture includes its problems. Delegated agents fail quietly: a sub-agent that misread the brief produces confident, wrong work, and a board that says “done” invites you not to check. Shared memory also means shared contamination — one bad instruction propagates to every session that inherits it. None of this argues against Claude Hub; it argues for treating agent output the way you treat a new contractor’s first fortnight, with review gates that relax as trust is earned. The teams that get value early will be the ones that design those gates before the feature arrives.
How Claude Hub Fits Anthropic's Product Arc
Seen in isolation, a settings toggle is trivia. Seen against two years of product history, Claude Hub is the visible tip of a deliberate arc — from static knowledge containers to autonomous, orchestrated work.
From Projects to Managed Agents
Anthropic launched Projects in June 2024 as knowledge containers: each project carries a 200K-token context window, the equivalent of a 500-page book. Managed Agents arrived on the developer platform in April 2026, giving API customers server-hosted agents with managed sandboxes. On 26 August 2026, Claude unified memory across Chat and Cowork for Free, Pro and Max users. Each step moves context and continuity out of the individual chat and into infrastructure.
The step Claude Hub completes
Managed Projects puts autonomous cloud sessions inside that continuity, and Claude Hub gives a human one surface to supervise them. The through-line is a repositioning of Claude as infrastructure for shared organisational work rather than solely a personal assistant. That is also why these leaks matter beyond Anthropic-watchers: they describe the shape of agentic work that natural language processing tools are converging on across the industry.
Claude Hub vs ChatGPT Projects vs Gemini AI Rooms
Every major lab is now building a “place where work lives,” and the three approaches differ more than the naming suggests. Google was spotted prototyping AI Rooms for Gemini Enterprise on 31 August 2026 — we covered what Rooms, Goals and Playbooks looked like in detail. OpenAI made ChatGPT Projects free for all users back in September 2025.
| Factor | Claude Hub + Managed Projects | ChatGPT Projects | Gemini AI Rooms |
|---|---|---|---|
| Status | Early Access flag, unannounced | Shipped, free tier since Sep 2025 | Prototype, spotted Aug 2026 |
| Core idea | Agent orchestration with a task board | Chat and file organisation | Shared rooms with Goals and Playbooks |
| Autonomy | Cloud sessions run without you | None — responds when asked | Playbook-driven, semi-guided |
| Collaboration | Shared sessions, threads, repositories | Single user | Team rooms, enterprise-first |
| File/context anchor | 200K-token projects, shared memory | 5 / 25 / 40 file caps by plan | Shared knowledge base |
| Likely first audience | Max, Team and Enterprise plans | Everyone, including free | Gemini Enterprise customers |
Three different bets
OpenAI’s bet is reach: ChatGPT reported one billion weekly active users in August 2026, up from 900 million in February, and Projects is a free organisational feature for that audience. Its file caps are 5 on Free, 25 on Plus and 40 on Pro. Google’s bet is the enterprise workspace, bundled where companies already buy. Anthropic’s bet — on the evidence of Claude Hub — is depth of autonomy: fewer users, doing heavier delegated work.
The commodity baseline
The uncomfortable read for Anthropic is that basic project organisation is already a commodity — free in ChatGPT for a year now. Claude Hub only justifies itself if orchestrated sub-agents produce outcomes a folder of chats cannot. That is a higher bar, and it is presumably why Anthropic is testing so deliberately rather than shipping a me-too folder view.
Who Gets Claude Hub First? Plans and Pricing Signals
Nothing about availability is confirmed, but the pricing page offers circumstantial evidence worth reading. Anthropic’s published plans, verified on 1 September 2026 at claude.com/pricing, are below.
| Plan | Price (monthly billing) | Relevant benefit |
|---|---|---|
| Free | $0 | Baseline chat |
| Pro | $20 ($17 billed annually) | Projects, more usage |
| Max | From $100 (5x or 20x usage) | “Early access to advanced Claude features” |
| Team | $25 per seat ($20 annual) | Collaboration, likely Managed Projects home |
| Enterprise | $20 per seat + usage | Deployment controls |
The Max plan clue
Max is the only consumer plan that sells early access by name, which makes it the natural first home for a Claude Hub Early Access cohort. The monthly prices above chart the gap plainly: a Max subscriber pays five times a Pro subscriber, and early features are part of what that buys.
Team and Enterprise signals
Managed Projects was reported in July as planned for Team and Enterprise plans, and TestingCatalog has separately noted that early-access organisations may already hold related Claude Code features. The plausible sequence is Max individuals testing Claude Hub while Team and Enterprise organisations pilot Managed Projects — but that is inference from plan design, not a confirmed rollout map.
What Your Business Should Do Before Claude Hub Ships
You cannot install any of this today, and that is precisely why now is the cheap moment to prepare. Orchestrated agents amplify whatever process discipline you already have — including the absence of it.
Prepare the work, not the tool
Agent orchestration presumes work that can be described as tasks with owners and done-states. If your projects live in heads and hallway chat, a task board run by Claude Hub will faithfully organise noise. Mapping which streams of work could be delegated — and writing down the instructions a competent contractor would need — is groundwork that pays off with any vendor’s agents. A structured AI strategy engagement is the fastest way to do that mapping honestly.
Governance questions to settle now
Managed Projects means cloud sessions acting autonomously on your files. Before that arrives, settle the boring questions: which data classes may enter a managed workspace, who reviews an agent’s output before it leaves the building, and how your cybersecurity team monitors long-running sessions nobody is watching in real time. Teams that already run intelligent automation have most of these answers; teams that do not should start with a pilot boundary, not a policy vacuum.
Watch the right signals
Ignore the hype cycle and watch three concrete tells: an official Anthropic announcement naming Claude Hub, the Managed Projects tile becoming tappable on mobile, and early-access pricing terms. Those three will tell you the rollout scope, the platform priority and the cost model — everything a budgeting decision needs.
What Is Still Unconfirmed
A responsible summary has to end with the hedges, because every load-bearing claim here is leak-based. Anthropic has published nothing about Claude Hub, Hub Mode or Managed Projects. There is no confirmed link between the mobile toggle and the Hub destination in Claude Code on the web, and enabling the flag has, at every reported stage, produced no operational change for the tester who flips it.
The gaps are specific and worth listing. The Early Access flag stage does not name a cohort, a date or a price. The mobile tile is an empty placeholder with no platform specified — the sighting does not even distinguish iOS from Android. The Orbit lineage is suggestive, not established. And a feature at the Early Access stage can still be reworked, renamed or cancelled; Orbit itself is the proof, having been absorbed or replaced before ever shipping under its own name.
What the record does support is narrower but still useful: Anthropic is investing sustained, staged effort in agent orchestration, the Claude Hub flag is moving in the direction of release, and Managed Projects is being fitted for mobile. If any of these harden into announcements, the calculus above changes from preparation to procurement — and we will cover it when it does.
References
TestingCatalog on Threads: Claude Hub transitions to Early Access, 1 September 2026
TestingCatalog on Threads: Anthropic keeps testing Hub internally, 29 August 2026
TestingCatalog: Anthropic tests Hub Mode in Claude for sub-agents
TestingCatalog: Anthropic develops Claude-driven Managed Projects
TestingCatalog: Anthropic develops its own Slack as Claude Code Projects
CryptoBriefing: Anthropic tests Managed Projects feature for Claude
Anthropic: Collaborate with Claude on Projects
TestingCatalog: Google develops AI Rooms for Gemini Enterprise
More AI coverage: explore Progressive Robot's AI Models, Tools & Releases hub — hands-on reviews, setup guides and benchmarks in one place.