Anthropic announced a redesigned Claude Code Projects beta on 2026-09-17 that introduces a coordinator agent directing parallel cloud-session threads, each with its own repository branch and copy, sharing a persistent project memory across threads. The beta is limited to select Pro/Max users on web/desktop (not CLI), with a promised wider Pro/Max rollout "over the coming week" that — per the case's timeline — is now ~3.5 weeks overdue with no observed landing. Vendor materials claim reduced manual delegation and handoffs for long-running, multi-repository work; independent field evidence accumulated since launch (direct beta reports, an adjacent orchestrator benchmark, and builders' parallel-session tooling) variously reports shared-memory degradation into a "telephone game," configured workers that never execute, and coordination friction (permission storms, cross-repo settings bleed) — while a separate 13-month usage log shows general Claude Code adoption reducing interruptions and enabling ~13 parallel projects/day. The snippets confirm the announcement and its core mechanics; they do not independently validate the coordination quality or the overdue rollout.
Anthropic's Projects beta is a consequential vendor attempt at the exact coordination problem Scott's frameworks argue requires durability-beats-coordination, a delegation plane with binding mechanisms, and deterministic control planes — not shared-memory orchestration. Field evidence (telephone-game degradation, workers-never-run benchmark, explicit-mechanism convergence by builders) now produces dated receipts for Scott's positions while the vendor's purpose-built coordinator remains unverified at scale. The overdue rollout deepens the window-closed risk, but the natural experiment has already yielded multiple independent lines validating Scott's architecture.
ip:framework.delegation-planeip:concept.durability-beats-coordinationip:concept.shared-blackboardip:framework.long-running-agentsip:concept.deterministic-coreip:framework.micro-agents-architectureip:concept.scatter-gather-cognitionradar:claude-code-session-messagingradar:anthropic-multiagent-patterns-validationradar:four-model-terminal-bench-backfireradar:agentlane-git-native-coordinationradar:deadlock-multi-agent-survival-benchmarkradar:civitas-persistent-agent-civilization
queries asked of Scott's wikis
- delegation-plane coordination binding mechanisms
- shared-blackboard memory prioritization accumulation
- durability-beats-coordination deterministic control plane
- agent-harnesses parallel session orchestration local vs cloud
- model-sovereignty local inference economics plan-limit burn
- long-running-orchestration verified completion fresh-worker recovery
| source | object | author | score | comments |
|---|
| 🟧 hn | Claude Code ProjectsRetrieved article excerptOpen article · Retrieved 2026-09-18T18:22:47.263793+00:00 # Projects redesigned: from folder to conversation
A new experience for Claude projects, now available in beta in Claude Code
- Category
[Product announcements](https://claude.com/blog/category/announcements)
[Claude Code](https://claude.com/blog/category/claude-code)
- Product
[Claude Code](https://claude.com/product/claude-code)
- Date
September 17, 2026
- Reading time
5
min
- Share
[Copy link](https://claude.com/blog/projects-redesigned)
https://claude.com/blog/projects-redesigned
Managing multiple sessions across a build used to require you to divide the work, juggle handoffs, and stitch the results back together. Now in a Claude Code project, you describe what needs to get done and Claude manages the work.
Claude scopes the request, delegates the work, coordinates parallel threads, reviews the outputs, and assembles the finished result. You can steer progress throughout, even from your phone, and it keeps working after you step away from your computer.
For example, configure a project and set a goal to reduce your app's checkout p75 latency. Then ask Claude to profile each endpoint, test optimizations, and open PRs in parallel threads. Or connect your API, web, and mobile repos and set a goal to retire a deprecated v1 endpoint. Claude creates a thread per repo to migrate the callers, run the tests, open PRs, and then tells you which ones need to merge first.
Starting today, updated projects are available in beta to select Claude Pro and Max subscribers who use cloud sessions in Claude Code and don’t have any existing projects on the web or desktop.
Over the coming week, we'll expand access to more Claude Code users on those plans. Updated projects across all of Claude and Team and Enterprise plans come after that. If you're on Pro or Max and don't have access yet, you can join the [waitlist](https://claude.com/form/projects).
Existing projects on Pro and Max plans keep working as they do today. We'll upgrade them as the rollout expands to chat and Cowork.
## Threads do the work, Claude directs it
Projects have threads that do the work and a coordinator that directs them.
When you start a project, you select a goal as well as the repo or context. Claude starts by suggesting work it can pick up right away. You can configure the project’s cloud environment, connectors, plugins, instructions, and model.
You can monitor and guide progress in the main project chat, or dive into each individual thread to examine and steer the details. Brief Claude in the project the way you'd brief a chief of staff and it routes work to new or pre-existing threads.
Claude also checks in and follows through on work. With repositories connected, a thread opens pull requests and runs your tests; with documents, it reads them and drafts.
Under the hood, each thread is a Claude Code cloud session working on its own branch and copy of the repo. The coordinator keeps work organized, but if any threads work on the same code, the overlap is resolved as a merge conflict just like any other PR.
Each thread can further split its delegated work into pieces using subagents, loops, and workflows when needed so large assignments finish faster.
## Context builds over time
Projects are designed for long-running or agentic workflows: work that takes longer than one reply and has more than one part.
Over time, Claude learns more about the project details and applies them to its work. Every thread now adds to and draws from a shared memory, reducing the need for complex prompt engineering.
For example, Claude can remember the release moved to Friday, why the export was dropped, or who to check in with before touching the billing service.
*One project across a week. Each request opens a thread on the right; the memory, the decisions, and the results build underneath.*
Claude also remembers your working and communication style. You can ask it to adjust how often it checks in, how frequently it starts new threads, or how detailed to make each update.
Alongside memory, projects now include a library that collects the files you add and the artifacts produced by Claude. This makes it easier to find relevant materials and for new work to build on past efforts.
## What's next
Projects can run several threads at once, and each one is a full Claude Code session. Because of this, projects can reach usage limits faster. You can check project specific usage and select the model and effort levels used by the coordinator chat as well as the worker threads.
Threads run in the cloud today; running on your machine alongside your local tools and code and behind your network is coming very soon.
Start using [projects](https://claude.com/projects).
No items found.
[Prev](https://claude.com/blog/projects-redesigned)Prev
0/5
[Next](https://claude.com/blog/projects-redesigned)Next
eBook
##
FAQ
No items found.
## Related posts
Explore more product news and best practices for teams building with Claude.
Jun 18, 2026
### Claude Code now supports artifacts
Product announcements
[Claude Code now supports artifacts](https://claude.com/blog/projects-redesigned)Claude Code now supports artifacts
[Claude Code now supports artifacts](https://claude.com/blog/artifacts-in-claude-code)Claude Code now supports artifacts
Jun 25, 2025
### Turn ideas into interactive AI-powered apps
Product announcements
[Turn ideas into interactive AI-powered apps](https://claude.com/blog/projects-redesigned)Turn ideas into interactive AI-powered apps
[Turn ideas into interactive AI-powered apps](https://claude.com/blog/build-artifacts)Turn ideas into interactive AI-powered apps
Jul 25, 2025
### Build and share AI-powered apps with Claude
Product announcements
[Build and share AI-powered apps with Claude](https://claude.com/blog/projects-redesigned)Build and share AI-powered apps with Claude
[Build and share AI-powered apps with Claude](https://claude.com/blog/claude-powered-artifacts)Build and share AI-powered apps with Claude
Jun 17, 2026
### Claude Design now stays on brand for daily work
Product announcements
[Claude Design now stays on brand for daily work](https://claude.com/blog/projects-redesigned)Claude Design now stays on brand for daily work
[Claude Design now stays on brand for daily work](https://claude.com/blog/claude-design-stays-on-brand-for-daily-work)Claude Design now stays on brand for daily work
## Transform how your organization operates with Claude
See pricing
[See pricing](https://claude.com/pricing#api)See pricing
Contact sales
[Contact sales](https://claude.com/contact-sales)Contact sales
Get the developer newsletter
Product updates, how-tos, community spotlights, and more. Delivered monthly to your inbox.
Thank you! You’re subscribed.
Sorry, there was a problem with your submission, please try again later. | smoser | 3 | 0 |
| 🟧 echo.blog ⭐ | Announces a limited Claude Code Projects beta with a coordinator, parallel cloud-session threads, shared memory, and continued execution aft | Anthropic | — | — |
| 🟧 hn | Anthropic's new Claude Code feature could drain your plan before lunch | Brajeshwar | 2 | 1 |
| 🟠 reddit | New Beta CC Feature: Projects coordinate work across multiple sessions from one conversation ClaudeAI | Wsz2020 | 57 | 25 |
| 🟠 reddit | Claude in Projects is definitely some novel engineering ClaudeAI | frubberism | 0 | 0 |
| 🟠 reddit | I benchmarked Claude Code orchestration against a solo session. My configured workers ran 0 times. ClaudeAI | itzxpringle | 3 | 13 |
| 🟠 reddit | Projects made Claude Code worse for me ClaudeAI | InternetWorker1 | 3 | 5 |
| 🟠 reddit | I run up to 6 Claude Code sessions on the same codebase. I built a control room so they stop stepping on each other (open source) ClaudeAI | Ramillax | 0 | 34 |
| 🟠 reddit | Parallel sessions: how do you stop the agent from treating what it loaded at session start as the current truth? ClaudeAI | NickolaiK | 1 | 21 |
| 🟧 hn | Claude Code in the Cloud | ilreb | 2 | 0 |
| 🟠 reddit | Do you like the unified "One Claude" experience so far? ClaudeAI | Balance- | 8 | 13 |
| 🟠 reddit | 13 months of Claude Code logs: I interrupt my agent 8x less, so I now juggle 13 projects a day (+ a survey for people who build with agents) ClaudeAI | anx3ous | 0 | 3 |
| 🟠 reddit | Not a fan of the Projects beta ClaudeAI | ShaktiExcess | 24 | 12 |
| 🟠 reddit | Tried the new Projects for Claude Code: best workflow I've had with it, held back by a handful of rough edges ClaudeAI | kbrianps | 1 | 3 |
2026-10-10T14:55:41Z
grounded: converges/high — Anthropic's Projects beta is a consequential vendor attempt at the exact coordination problem Scott's frameworks argue requires durability-beats-coordination, a
2026-10-10T14:38:46Z
First-hand Projects beta workflow report (1x2ez3d) added: user calls it "best workflow I've had" but documents auto-mode permission storms and cross-repo settings bleed — independent corroboration of shared-memory coordination friction. The wider Pro/Max rollout remains ~3.5 weeks past its pledged "coming week" with no observed landing, deepening window-closed risk. The natural experiment continues producing receipts for Scott's delegation-plane / durability-beats-coordination / deterministic-control-plane positions while the vendor's purpose-built coordinator stays unverified at scale.
2026-10-10T13:39:44Z
evidence attached: reddit.post.1x2ez3d — First-hand user report on Claude Code Projects beta workflow (auto-mode permission storms, cross-repo settings bleed) — independent corroboration of the coordinated-cloud-threads episode.
2026-10-10T00:33:31Z
New field evidence adds a positive long-term usage signal (13-month log analysis: reduced interruptions, 13 parallel projects/day) but also direct negative feedback on the Projects beta itself (memory leaks between threads, single-session friction for non-coding work). The wider Pro/Max rollout remains ~3 weeks past its pledged 'coming week' with no observed landing, deepening the window-closed risk. The case's meaning sharpens: the natural experiment continues producing receipts for Scott's frameworks (delegation-plane, durability-beats-coordination, deterministic-control-plane) while the vendor's purpose-built coordinator remains unverified at scale.
2026-10-09T19:58:38Z
evidence attached: reddit.post.1x1m24k — Direct user feedback on Projects beta redesign: memory leaks between threads, single-session friction for non-coding work — adoption evidence for the coordinated-cloud-threads hypothesis.
2026-10-09T19:58:38Z
evidence attached: reddit.post.1x1ocdn — Builder's 13-month log analysis shows reduced interruptions and 13 parallel projects/day — direct user evidence on the Projects beta's multi-project orchestration claim.
2026-10-08T08:41:56Z
The "One Claude" unified experience rollout appears underway per one user report, but the Projects beta's core coordination claims remain unverified and the wider Pro/Max access pledge is ~3 weeks overdue. The case continues to trend toward window-closed absent decisive field data on the purpose-built coordinator.
2026-10-08T08:40:10Z
evidence attached: reddit.post.1x0ljil — User describes Anthropic's rolled-out 'One Claude' unified experience (merged Chat/Cowork, cloud persistence, cross-device skills/plugins) — user feedback on the same coordinated-cloud-sessions rollout the case tracks.
2026-10-06T17:44:58Z
The attached 'Claude Code in the Cloud' HN item is vendor-side documentation of the already-known cloud-session surface — no inspectable independent content, 2 points / 0 comments — so it adds no new evidence line and nothing about the case's meaning changes: need validated, mechanism contested, the beta's core claims still unverified. The wider-rollout anomaly deepens (~3 weeks past the pledged 'coming week'); the case now lives or dies on whether decisive purpose-built-coordinator field data or a confirmed quiet stall arrives.
2026-10-06T16:42:15Z
evidence attached: hn.story.49980091 — Anthropic-side claude.dev blog on running Claude Code in the cloud is directly the surface of the coordinated cloud-sessions episode.
2026-09-28T20:09:32Z
New independent lines (NickolaiK's stale-state thread; Ramillax's thread growing 11→24 comments with other builders' tools surfacing) consolidate one pattern — parallel sessions collide and builders converge on enforcing mechanisms over shared context, distilled as 'Rules alone didn't hold, mechanisms did' — which crosses the two-independent-lines bar for the need-validated/mechanism-contested reading; the state moves to corroborated on the belief side even though nothing new emerged about the beta itself. Heat stays low (0.17 pts/h vs a 24 pt/h peak; additions repeat known ground, hence no material change), but note the promised wider Pro/Max access is now well past its 'coming week' with no observed landing — watch for either the decisive field-data wave or a quiet stall.
2026-09-28T18:36:53Z
evidence attached: reddit.post.1wslfka — Documents the exact parallel-session stale-state and cross-session coordination failure mode that the Projects beta claims to solve, with working mitigations.
2026-09-28T03:34:45Z
grounded: converges/high — The natural experiment is now producing receipts in Scott's favor: InternetWorker1's shared-context 'telephone game' is the serial-translation/knowledge-graveya
2026-09-28T03:26:34Z
The control-room project reframes the case from 'contested claim' to 'validated need, contested mechanism': an independent builder confirms parallel multi-session Claude Code coordination is a real, painful workflow worth custom tooling, and its explicit visibility/write-conflict design stands as the rival to the beta's shared-memory approach — the only mechanism with direct field evidence, which remains negative (telephone-game report, workers-run-zero-times benchmark). This is a belief-side addition, not an attention one: periphery is static and absolute rates are cold (0.5 pts/h; the 77th-percentile reading is an artifact of an aged cohort, not live spread).
2026-09-28T03:25:10Z
evidence attached: reddit.post.1ws30pa — Independent open-source control room for parallel Claude Code sessions on one codebase shows the multi-session coordination need Anthropic's Projects beta targets is real and unsolved for local runs.
2026-09-24T15:28:08Z
First direct hands-on of the actual beta (InternetWorker1) reports shared project memory re-internalizing all context with no prioritization — an 'endless telephone game' — so both real-world lines (adjacent orchestrator benchmark, direct field report) now point against the coordination claim instead of merely leaving it unverified. Engagement is quiet (0.83 pts/h, static periphery despite hot sibling concepts), so this is a belief update, not an attention one.
2026-09-24T15:25:48Z
evidence attached: reddit.post.1wp3fyy — Negative field report that Projects' shared context degrades into a telephone game directly counters the product's coordination claim.
2026-09-24T14:34:46Z
First hands-on evidence arrives, and it cuts against the premise: a self-built orchestrator benchmark where configured workers ran zero times and multi-agent setups matched but never beat solo Opus 5.5 moves the case from 'consequential beta awaiting evidence' to 'consequential beta whose coordination premise now faces direct adjacent counter-evidence' — whether delegation instructions actually bind is now the crux. The announcement surge has fully cooled (0.17 pts/h, 33rd percentile, periphery not expanding), so the fresh evidence earns a material flag, not attention temperature.
2026-09-24T13:28:09Z
evidence attached: reddit.post.1wp119w — Contrarian benchmark: orchestrated multi-worker setups matched but never beat a solo Opus 5.5 session and configured workers ran zero times — material counter-evidence on coordination value.
2026-09-21T14:14:11Z
relevance=high case never alerted; deterministic escalation to deliver
2026-09-21T12:22:31Z
evidence attached: reddit.post.1wmahgg — The leaked Projects documentation materially contextualizes how Anthropic's coordinated cloud-thread workflow is implemented.
2026-09-19T00:30:35Z
The newly attached Reddit post quotes Anthropic’s newsletter rather than reporting hands-on use, so it does not supply the independent corroboration claimed by the attachment rationale. This remains a relevant beta to evaluate, but repeated announcement coverage adds no evidence of reliable coordination or reduced handoffs.
2026-09-19T00:22:10Z
evidence attached: reddit.post.1wk6vp4 — Independent user report corroborates the new multi-session, cloud-running Projects workflow.
2026-09-18T21:07:36Z
The attached coverage does not independently validate orchestration quality or reduced handoffs; its usage-cost framing largely repeats Anthropic’s own warning. This remains a concrete beta worth evaluating against Scott’s delegation and shared-memory work, rather than a demonstrated solution to durable multi-repository execution.
2026-09-18T20:22:41Z
evidence attached: hn.story.49759033 — Independent coverage of Claude Code's parallel-projects feature bears directly on the open case about coordinated persistent cloud coding sessions.
2026-09-18T18:35:48Z
grounded: converges/high — Anthropic’s coordinator and shared project memory bring a consequential vendor toward Scott’s Delegation Plane and Shared Blackboard positions inside Claude Cod
2026-09-18T18:30:24Z
case created — The first-party beta announcement establishes a distinct orchestration release beyond existing chat unification and self-hosted execution cases.