digest
Short Summary
Main recommendation: Use Claude Code Projects for large, multi-session goals by giving Claude one overarching goal and letting it split the work into parallel threads, each running as its own cloud session on its own branch. The video demonstrates this with a website performance project that ends in an artifact report and a daily regression-check routine.
Key Points
- A project is one ongoing conversation where Claude coordinates work, and each task gets its own thread, which keeps running after you close your laptop and can be checked from your phone.
- Projects suit goals spanning many sessions or multiple repos, such as migrating away from a deprecated package or updating an endpoint across API, web, and mobile repos.
- Threads can be switched to run on your local computer when a task needs local files or tools.
- Threads handle the work: they open pull requests, watch them, push fixes when CI fails, and address review comments.
- Artifacts and shared Library files give the team a persistent place for reports and specs that Claude can update later.
Details Worth Noting
- Concrete instructions:
- Open the Code tab on desktop or the web app, create a new project, and enter a project name, a goal for Claude, and the relevant repos or folders.
- In the main conversation, share metrics and user feedback and ask Claude to explore the codebase and suggest improvements; after you agree on tasks, Claude spins up a thread per task.
- Click a thread card to follow progress, steer by talking in the thread, or stop it; check the Threads tab for threads waiting on you.
- To review PRs, open the Pull Request tab in Overview and ask Claude when you are ready to merge.
- To prevent regressions, connect Claude to your monitoring service, then ask it to set up a daily routine that checks real user metrics and opens a PR if it finds a regression.
- Usage tips: set the coordinator to low effort, since it mainly routes work; consider a smaller default model and switch to a larger one within threads for complex tasks; and tell Claude preferences such as how many threads to run at once, update frequency, or whether to show a plan first, storing them in project memory that all threads share.
- Projects use plan limits faster than a single session because each thread is a full code session that can run in parallel.
- Projects are in beta and rolling out gradually to Pro and Max plans; the video points to linked docs for details.
Bottom Line
Projects are best for ambitious goals too big for one session, and their value depends on managing usage through effort settings and model choice.
