Anthropic’s new Projects in Claude Code replace the old folder‑based project with a single ongoing conversation that acts as a coordinator and launches parallel cloud threads for each piece of work. While this gives you true parallelism and persistent background execution, it also creates practical challenges that teams need to address head‑on.
First, the cost model changes: every running thread consumes a full session of your plan, so a project can burn through limits much faster than a single chat. Teams should start by setting separate effort levels—low for the coordinator, high for threads—and regularly check the per‑project Usage tab to spot unexpected spikes. Idle threads still wake on CI failures or review comments, so consider closing threads once their pull request is merged or marking them as idle when work stalls.
Second, overlap between threads surfaces as ordinary git merge conflicts. To minimize friction, break work into clearly bounded scopes before asking Claude to spawn a thread, and use the coordinator to review thread reports frequently. When two threads must touch the same file, let the coordinator mediate by having one thread open a draft PR and the other comment on it, reducing surprise conflicts.
Third, shared context is vital but easy to overlook. All threads automatically inherit the project’s repositories, uploaded files, project instructions (up to 16 000 characters) and the MEMORY.md index. Keep MEMORY.md up‑to‑date with decisions like release dates or stakeholder contacts, and avoid duplicating that information in each thread’s prompts.
Fourth, permission rules, hooks and environment variables apply only from the directory where a thread starts. In multi‑repository projects this means a thread may not inherit repo‑level hooks; place any required setup in a shared setup script that each thread calls at startup, or rely on the coordinator to enforce global policies.
Finally, stay within the 200 new threads per day ceiling by batching similar tasks into fewer threads or using subagents and loops inside a thread for finer granularity. Use the Overview pane to filter threads by state (Ready for review, Waiting on you, Working, Landing, Idle, Resolved) and the Library tab to track generated files.
By defining clear scopes, monitoring usage, keeping shared memory fresh, and standardizing thread initialization, teams can harness the parallel power of Claude Code Projects without drowning in cost, conflicts, or confusion.
#AI #Product #ClaudeCode #DevOps #Automation #ML