ZDNET’s key takeaways
- Claude Code Projects could tame multi-agent chaos.
- Cloud users get persistent threads and shared memory.
- Local workflows remain outside the new system for now.
Claude Code has always had projects, in the sense that a certain directory is declared as root for source code. I keep each of the apps I develop in separate directories, and I’ve built a complex orchestration and management workflow that allows multiple messages to be exchanged between these directories. Beyond that, they are isolated code structures.
Also: After the vibe-coding rush comes the debugging hangover
Today, Anthropic announces that it is reshaping the idea of projects in the Claude Code, adding both superpowers and hard limits. For certain classes of work, they can be game changing. For other types of work, they are essentially irrelevant.
Currently, this capability is offered to “Select Claude Pro and Max subscribers who use cloud sessions in Claude Code and do not have existing projects on the web and desktop,” Anthropic said. The features will roll out to other users over time, but the company said that “If you are a Pro and Max subscriber already using Projects, they will remain as they are for now, and we will upgrade them later.”
That rules me out, because all my development related to the Claude code is done in the terminal and on local folders. But for those who live in the cloud, the new project capability brings these structures to life.
Anthropic said: “Claude scopes the request, delegates the work, coordinates parallel threads, checks the outputs, and assembles the finished result. You can control the progress, and it continues to work after you close your laptop.”
Features for cloud coding
One of the more pressing challenges when starting a bunch of agents is keeping them all on track. Imagine trying to get 20 oh-so-fluffy kittens to walk through a small door in alphabetical order by their kitty names. It won’t happen, but it will be a lovely overload of coordination errors.
Also: Treat your AI agents like fearful but misguided human interns – before you lose control
Coding agents are like that, but much less cute. To get agents to work together on a goal, you need orchestration. This is basically a manager-level coordinator who goes beyond the prompt. Claude Code’s revamped projects feature provides for that orchestra.
You also get addressable threads. Each thread becomes a full cloud session, with its own branch and copy of the repo it’s working on. The coordinator provides instructions that route sessions and existing threads as appropriate.
Anthropic advised you to inform Claude about a project “as you would inform a chief of staff, and it will lead to work on new or pre-existing threads.”
Because these threads are named, you can steer them in mid-flight. You can grab into a running thread and redirect it as needed. No more waiting for it to finish, wasting your time and consuming tokens, only to produce something that you can already see going wrong.
Also: How to install and configure Claude Code, step by step
Threads can be aware of each other. Anthropic said: “Over time, Claude learns more about the details of the project and applies them to his work. Each thread now adds to and pulls from a shared memory, reducing the need for complex prompt engineering.”
This is a form of automated memory that can prove quite powerful in the right context. It allows your work to span repos, so you can work across multiple repos at once, working with one thread attached to each, and then get feedback on which pull requests need to be merged and in what order.
And because it’s all in the cloud, when your computer is turned off, you can keep it running. Anthropic added that mobile check-ins are on the way.
Which is not new
This is not a new parallelism. I’ve been using swarms of subagents for a long time. They can find out and work. In fact, Anthropic’s announcement stated that this architecture is what the new projects will use internally.
Also: 75% of the developers I asked prefer Claude Code over Codex – here’s why
Claude also divided the work between sub-agents. The difference is that the split persists, and you can go back to the discussion for each subagent and chat about it.
Another thing to note: this feature doesn’t seem to be strictly autonomous. It looks like threads start when you press the start button for each thread that Claude recommends.
As mentioned above, at least for now, this is not yet a tool for local developers like me. But Anthropic added that “Threads are running in the cloud today; running on your machine alongside your local tools and code and behind your network is coming very soon.”
Also, Anthropic does not remove the terminal. This is an additional feature. Nothing the company described in its blog post indicates any adverse changes to local or CLI functionality.
Put everything in perspective
While I was writing this, I stopped and ate a burger. While munching, I thought about what this new capability means and came up with three ideas worth considering. Here’s a quick rundown of things to consider when evaluating this skill.
- It remembers, but what does that mean? Memory is not well documented in Anthropic’s post. There is no apparent way to check what is saved, correct it if it goes wrong, export it, nor any details about how long it is kept.
- Cost and usage scale as you roll out more agents. This is now the case, with subagents, but it is worth reinforcing. Anthropic said: “Projects can run multiple threads at once, and each one is a full Claude code session. Therefore, projects reach usage limits faster.” I’m definitely cut off after running a fleet of agents over my code.
- Conflicts are pushed to the streets; they are not prevented. Threads that overlap will merge as conflicts, which you need to resolve manually.
Questions for Anthropic
I have many, many, oh-so-many questions for Anthropic about this, especially if some of these capabilities are available to desktop users. I’ve narrowed them down to three that are worth thinking about, at least for now.
- When a project outgrows its context (which mine does every few hours or so), what happens to that shared memory? Can a user see what has been kept? Are they all available as .md files somewhere?
- Is the history of a project automatically kept, and can it be exported with intact prompts? My memory management system requires Claude to record every prompt I give it, which provides both an engineering audit trail and something of a scientific journal for evidence of conception, diligence, reduction of practice and ingenuity.
- If the ability to use this feature is available on local projects, will projects inherit memory, permissions, and hook behavior? How does it work with existing memory management systems and existing, deeply evolved projects?
From Anthropic’s specific statement that this feature is for those who “don’t have existing projects on the web and desktop,” I’m guessing the migration isn’t resolved yet.
If you could control a running Claude code thread before it goes off the rails, how much would that change how you use coding agents? Let us know in the comments below.
You can follow my daily project updates on social media. Be sure to subscribe my weekly update newsletterand follow me on Twitter / X at @DavidGewirtzon Facebook at Facebook.com/DavidGewirtzon Instagram at Instagram.com/DavidGewirtzon Bluesky at @DavidGewirtz.comand on YouTube at YouTube.com/DavidGewirtzTV.
See full bio