Anthropic Turns Claude Code Into a Multi-Agent Team With New Projects Beta

Anthropic relaunched Projects inside Claude Code on September 17, giving developers a coordinating agent that splits a single engineering goal into multiple parallel Claude Code sessions, called threads, each running as a cloud session on Anthropic's servers with its own branch and repository copy. The beta is rolling out first to select Claude Pro and Max subscribers, with no local execution yet and no access for Team and Enterprise customers, and The New Stack reported the parallel-projects feature could drain a subscriber's usage. Anthropic's own engineering team previously demonstrated the coordination approach by building a C compiler with a team of parallel Claude instances.

Anthropic just handed Claude Code a coordinator that splits engineering goals into parallel cloud agents, and it's rolling out now to select Pro and Max subscribers. Anthropic relaunched Projects inside Claude Code on September 17, giving developers a coordinating agent that can break a larger engineering goal into smaller jobs and hand them to multiple Claude Code sessions running at once. According to VentureBeat, each of those sessions is called a thread, and every thread is a cloud session: Claude Code running on Anthropic's servers rather than on your laptop, working its own branch and its own copy of the repository. The idea is simple enough. You give the coordinator a goal, not a task list, and it figures out how to divide the work. One thread might refactor an API. Another might update the tests that touch it. A third might write the docs. Each one runs independently, and each can spin up its own subagents, loops, and workflows if the job it's been handed turns out to be bigger than expected. That independence is also where things get messy, and Anthropic doesn't pretend otherwise. When two threads touch the same file, the conflict resolves the way it would for any two developers pushing to the same branch: like a pull request that needs a merge. Claude Code's own documentation describes the thread's card in the conversation surfacing a button for whatever comes next, resolve conflicts, fix CI, address comments, or merge it. It's not magic. It's git, with an AI on both ends of the disagreement. Shared memory is what separates this from just running five terminal windows at once. Every thread reads from and writes back to the same Project memory, so a decision made in one thread, a release date that moved, a feature that got cut, an instruction about who to check with before touching a particular service, stays visible to every other thread working the project. That's the part that used to be entirely manual. Developers who wanted parallel Claude sessions had to divide the work themselves and stitch the results back together by hand. Anthropic Resets Claude Code Usage Limits Again After a Rough Week of Outages https://startupfortune.com/anthropic-resets-claude-code-usage-limits-again-after-a-rough-week-of-outages/ Anthropic quietly reset Claude Code's 5-hour and weekly usage limits for every user on July 10, the latest in a string of resets following a rough stretch of outages. The wipe lands just three days before a temporary 50% weekly limit boost, widely read as a response to OpenAI's Codex, is set to expire. - claude code usage limits reset https://startupfortune.com/anthropic-resets-claude-code-usage-limits-again-after-a-rough-week-of-outages/ - anthropic outage compensation july 2026 https://startupfortune.com/anthropic-resets-claude-code-usage-limits-again-after-a-rough-week-of-outages/ Anthropic's own engineering team has already shown what that looks like at scale. In a recent write-up on its site, the company described building a C compiler with a team of parallel Claude instances working together, an internal demonstration of the exact coordination problem Projects is now trying to solve for outside developers. Cloud-only, and a narrow beta At launch, every thread runs in the cloud. There's no local execution yet, though Anthropic says local tool support is coming. That's a real limitation for developers who keep sensitive code or credentials off Anthropic's infrastructure, and it draws a hard line around who can use this today. The beta is narrow by design, too. It's rolling out first to select Claude Pro and Max subscribers, according to VentureBeat, and Team and Enterprise customers don't have it yet. Anthropic is testing this on individual developers before it touches organizations where a runaway coordinator could rack up a very different kind of bill. That bill is worth taking seriously. The New Stack reported that the new parallel-projects feature could drain a subscriber's usage plan fast, since spinning up several cloud threads at once multiplies token consumption in a way a single conversation never did. Frankly, that's a tradeoff every team adopting this will need to run the numbers on before letting a coordinator loose on a real sprint. Anthropic versus OpenAI, again The competitive read here isn't subtle. OpenAI has spent the past year building out Codex's own cloud agent tooling, and Anthropic's move pushes Claude Code further from an assistant that writes code next to you and toward a platform that runs a shift of AI developers while you sleep. Whichever company actually gets multi-agent coordination reliable enough for production codebases, not demo repos, stands to own a meaningful share of how software gets built over the next few years. For now, Projects is a beta with real limits: cloud-only execution, a narrow subscriber list, and a token bill that can climb faster than a developer expects. But the direction is clear. Anthropic isn't just improving Claude Code's ability to write a function. It's trying to make it manage a team. Also read: CoreWeave Plans Another $3.5 Billion In Convertible Notes To Fund AI Growth https://startupfortune.com/coreweave-plans-another-35-billion-in-convertible-notes-to-fund-ai-growth/ • Sword Health Is Buying Headspace for Up to $300 Million in Cash https://startupfortune.com/sword-health-is-buying-headspace-for-up-to-300-million-in-cash/ • Marc Benioff Tells AI Companies to Regulate Themselves or Face Lawsuits https://startupfortune.com/marc-benioff-tells-ai-companies-to-regulate-themselves-or-face-lawsuits/ Hackers Are Draining Claude Subscribers Usage Without Stealing a Password https://startupfortune.com/hackers-are-draining-claude-subscribers-usage-without-stealing-a-password/ Anthropic has confirmed that infostealer malware, including Vidar, LummaC2, and StealC, is being used to hijack Claude subscribers' login sessions and drain their paid usage without ever needing a password. The company has refunded some victims but still can't show users an itemized log of what was stolen or how. - how to protect Claude account from infostealer malware https://startupfortune.com/hackers-are-draining-claude-subscribers-usage-without-stealing-a-password/ - AI subscription accounts getting hacked without password theft https://startupfortune.com/hackers-are-draining-claude-subscribers-usage-without-stealing-a-password/ This article is posted in AI News https://startupfortune.com/category/ai/ , check it out for more related stories. Join the discussion Open in the community → https://startupfortune.com/community/ Almost there. Sign in and your reply posts straight away.