September 20, 2026
On September 17, 2026, Anthropic launched Claude Code Projects, and VentureBeat's writeup describes something genuinely new in how AI writes software: a single, always-on conversation that acts as a coordinator for your engineering work. You talk to it in natural language about what you're trying to get done, and Claude decides which pieces of that work deserve their own thread. Each thread then runs as a full Claude Code cloud session, on its own git branch, working against its own copy of the repository. Threads run in parallel, keep working after you've closed your laptop, and report back to the coordinating conversation -- which retains memory of the project over time. It's rolling out first to select Claude Pro and Max subscribers; Team and Enterprise customers don't have it yet.
It's easy for anyone selling a competing product to wave this off, so let's not. Coordinating multiple parallel, long-running agentic sessions across separate branches and separate repo copies, with a persistent conversation that remembers the state of the project and delegates work without you manually spinning up each session -- that's a hard orchestration problem, and Anthropic clearly solved a meaningful chunk of it. Most "AI coding agent" launches are still single-threaded: one prompt, one session, one diff. Claude Code Projects is describing something closer to how a technical lead actually runs a sprint -- breaking work into parallel tracks, letting each track run unsupervised, and checking in on progress. That's a fair and positive thing to say about this launch, full stop.
Read the mechanics again: threads on their own git branch, their own copy of the repository, reporting back into a coordinating conversation. Every one of those nouns -- branch, repo, coordinating session -- assumes you already have a codebase, already think in version control, and already have someone on staff who can look at a pull request and decide whether to merge it. Claude Code Projects doesn't remove the need for an engineer. It makes an engineer, or a team of them, dramatically faster at work they already know how to scope, structure, and review. That's a compounding tool for people who already write code -- which is exactly why it's rolling out to Pro and Max subscribers first, the individual developers and technical builders who live inside Claude Code every day, rather than to Team and Enterprise accounts where non-technical stakeholders would be the ones opening the conversation.
When a small or mid-market business says "we should just build this ourselves," they are almost never sitting on an existing repository with a branch strategy and a reviewer on staff. They're usually an operations person, an office manager, or a founder who knows precisely what the internal tool needs to do -- track inventory across three locations, route support tickets, manage appointment reminders -- and has no one who can turn that into code, let alone review someone else's pull request. Handing that person Claude Code Projects doesn't solve their problem; it hands them a more powerful version of a tool they still can't operate, because the missing piece was never orchestration speed. The missing piece was needing to write and review code at all. That's the same gap we've written about before when looking at tools built for engineers versus tools built to remove the engineer requirement entirely, including Factory's approach to AI coding agents and the broader trend of companies choosing to build instead of buy without fully reckoning with what that choice requires of them.
This is why Claude Code Projects and something like ViibeStack's AI app builder aren't really chasing the same customer, even though both get described as "AI-built software." Claude Code Projects compounds the leverage of people who already know how to write and review code -- it takes a team that could ship two features a week and gives them a shot at shipping five, by letting them run parallel workstreams without babysitting each one. ViibeStack is built for the opposite starting condition: no repo, no branch strategy, no engineer on staff, just an operations person describing what they need in plain language and getting a working, editable internal tool back, no pull request required. One tool multiplies engineering capacity. The other removes the requirement to have engineering capacity in the first place. Those are different problems, and a business evaluating build versus buy should figure out which one they actually have before picking a tool -- a question we walk through in more detail on our buy vs. build vs. ViibeStack comparison.
Ask it plainly: does your business already have a codebase, a git workflow, and someone who reviews merges before they go live? If yes, tools like Claude Code Projects are exactly the kind of leverage you should be adopting, and you should be excited about it -- this is a legitimate acceleration for technical teams. If the honest answer is no -- if what you actually have is a person who understands the operational problem cold and no one to build the fix -- then the coordination layer isn't your bottleneck, and no amount of parallel branches changes that. That's the gap tools like ViibeStack's internal tools software and workflow automation products are built to close, and it's worth being clear-eyed that Anthropic's launch, impressive as it is, was never aimed at closing it.
Sources