If you’ve used Claude Code, Codex, or other coding agents, you’ll soon face the next challenge:
AI can handle many tasks at once, but in the end, you end up managing a bunch of AIs yourself.
One session fixes the API.
Another session changes the web.
A third session runs tests.
Not only do you have to remember what each agent has done, but you also need to repeat requirements, handle dependencies, and finally piece the results back together.
On September 17, Anthropic launched a redesigned Claude Code Projects to solve this problem.
It’s not just adding another folder inside Claude Code; it places a Coordinator layer over multiple coding agents.
What is Claude Code Projects?
The simplest way to understand it:
Previously, you managed many Claude Code sessions yourself; now you give the entire project to one Claude, which then manages the multiple sessions beneath it.
When you create a Project, you provide:
- The project goal
- The repository or other context
- Cloud environment
- Connectors
- Plugins
- Instructions
- The model used
Then, instead of dumping everything into a single conversation, Claude first understands the overall goal and decides which tasks should be broken into different threads.
The official explanation is straightforward:
Threads do the work; Claude directs it.
Meaning:
The threads are responsible for the tasks.
The Claude Coordinator oversees the division and coordination.
A single requirement can be split into several parallel tasks
For example, your website’s checkout process slows down.
Previously, you might:
Ask AI to find the issue first.
Wait for it to finish, then test the API.
Then work on the database.
Then update the frontend.
Finally, run tests.
Claude Code Projects can break the work apart.
For example:
- Thread A: Analyze endpoint
- Thread B: Identify database bottleneck
- Thread C: Test various optimizations
- Thread D: Prepare pull requests
These tasks can advance simultaneously.
You remain in the top-level project conversation, monitoring overall progress, and dive into any thread as needed.
Each thread edits separate code, not the same code causing chaos
This is very important.
In Claude Code Projects, each work thread is actually an independent Claude Code Cloud session.
When working on a repository, each thread has its own:
- Branch
- Repository copy
- Working context
This allows multiple threads to proceed in parallel.
If connected to a repository, threads can also:
- Modify code
- Run tests
- Create pull requests
But this doesn’t mean agent conflicts disappear.
If two threads simultaneously edit overlapping code, normal merge conflicts can still occur.
In other words:
The coordinator manages the work but doesn’t eliminate Git conflicts.
The real innovation: Shared Memory
Parallel agents aren’t the most unique aspect.
The truly notable feature is the Project Memory.
Each thread not only completes its own work but important decisions and context generated can enter a shared memory for the entire project.
For example, if you tell Claude:
“Change the release from Wednesday to Friday.”
“Cancel the export feature for this version.”
“Confirm with a specific engineer before modifying the billing service.”
Other threads working on related tasks can utilize this accumulated project context.
The biggest difference from usual chat is:
Previously, you saved conversations; now you save a continually evolving work state.
Also included: A library to store documents and artifacts within the project
Claude Code Projects introduces a library.
Files added to the project and artifacts created during Claude’s work remain here.
New tasks can build upon these materials.
The overall architecture becomes:
Goal
→ Coordinator
→ Multiple Threads
→ Shared Memory
→ Files / Artifacts
→ Ongoing Use for Future Work
This model is closer to a true long-term software project rather than a single Q&A chat.
How it differs from the original Claude Projects
The name is easier to confuse.
The original Claude Projects was more like:
Combining chats, files, instructions, and knowledge in a single workspace.
The new Claude Code Projects goes further:
Not only saving the context you give Claude but also coordinating multiple active work threads.
To simplify:
Claude Projects helps AI remember "what you told it."
Claude Code Projects helps AI remember "what the entire project is doing."
Two current limitations to be aware of
First, not everyone has access today.
Anthropic currently offers the beta only to some Claude Pro and Max users who already use Claude Code Cloud sessions, with eligibility requirements at launch.
The company states they will gradually expand Pro/Max access over the next week before extending to Claude, Team, and Enterprise users.
If you don’t see it yet in your account, the feature isn’t absent.
Second, threads currently run in the cloud.
This means the work can continue even when your computer is off.
On the flip side, cloud threads can’t yet directly use your local tools, local code, or internal enterprise environments.
Anthropic will add local support later.
A practical cost consideration
A project can run many threads simultaneously.
Sounds convenient, but each thread is a full Claude Code session.
So:
The more parallel agents you run, the faster your usage quota may be consumed.
Anthropic offers project-specific usage controls, letting you select model and effort levels separately for the coordinator and worker threads.
So don’t assume "lots of agents" means "the better."
The real question is:
Does the task truly need parallel processing?
Claude Code Projects truly changes "who manages AI"
Previously, the competition among coding agents was mainly:
Who writes code faster?
Who is better at debugging?
Who can build bigger features?
Now, the question is moving up a level.
When a person uses three, five, or more agents at once, the real bottlenecks become:
- Who breaks down the work?
- Who remembers dependencies?
- Who knows which parts have changed?
- Who integrates the results?
- Where is human intervention needed?
Claude Code Projects’ answer is:
Put another AI on top to manage the AIs for you.
This may be the next important step for coding agents:
From AI Programmer,
To AI Team Coordinator.
Today, make a little progress with AI.
Learn one AI skill each day.
Save a little time every day.
Improve a bit of your capability every day.
SasaDaily grows with you.
