💥NEW AI CODING TUTORIAL: How To Start Claude Code Projects The Right Way (@claudeai ) TIMESTAMPS 00:00 Don't make the same mistake I did 02:30 Phase 1: PLAN 02:54 2 Questions to ask before starting 04:44 How to use AI to help you plan 06:14 Creating a project spec doc 11:36 Phase 2: SETUP 12:04 GitHub Repo setup 13:32 Create your environment variable file (.env) 13:50 CLAUDE md (and what to put in it) 15:57 Automated Project Documentation 18:24 Install Plugins 19:33 Install MCP Servers 20:52 Setup Custom Slash Commands and Sub-agents 23:22 Advanced setup: Preconfigure Permissions 24:09 Advanced setup: Hooks 24:55 Phase 3: BUILD 25:41 Building Your MVP with Claude 26:33 Workflow 1: Single Feature Development 27:49 Workflow 2: Issue based development 29:19 Workflow 3: Multi-Agent Development (Multi-Clauding) 30:39 Tips for Building Productively 33:04 Applying what you learned

"I usually toss 10 to 15 tasks to the Project, let it cook, and come back to verify screenshots or steer the threads as needed." -@pranathiperii (designer on the Claude Code team) on Claude Projects. More on working in the background, and why cloud agents transform what's possible with Claude Code:

Very impressed by Opus 5.5, it's now my daily driver (along with GPT-6 Astra)

Claude Code Projects is an EXTREMELY powerful new feature. But getting good results from it requires a different way of working with coding agents than you might be used to. I call it the "Chief of Staff" paradigm. Before: You drive Claude. You tell it what you want and how you want it done (maybe with some sophisticated prompt engineering). After: You brief Claude like you would a Chief of Staff, with high level goals and outcomes, and Claude drives a team of agents to achieve your goals. I explain the Chief of Staff paradigm of working with Claude Code below:

Projects introduces two types of agents in Claude Code: (1) a coordinator and (2) a thread. The coordinator agents is your chief of staff which managed all the threads in a Project. Threads are where the work happens. They are run copies of your repo in isolated sandboxes. The coordinator relays instructions but you can also talk to threads directly. I demo interacting with the coordinator and threads here:

Using the new Projects feature in Claude Code simple: Give Claude a task in plain english. The coordinator agent will spin up threads to plan, execute, and verify the work.

💥 NEW TUTORIAL: Claude Code Projects Clearly Explained in 39 Minutes Anthropic just dropped @claudeai Projects, a new paradigm for building with Claude Code where you just tell Claude everything you want it to do at once, and Claude will spin up multiple agents in the cloud to make it happen. I think using Claude Projects makes it 10X easier to build with Claude, but it also offers us a glimpse into the future of coding agents. In this video, I explain what Claude Projects are and how they work, how to use them well (with advice from the Anthropic team), plus demos of Claude Projects in action on real applications in my business. TIMESTAMPS 0:00 Claude Projects make building 10x easier 0:55 What are Claude Projects and how do they work? 3:30 Old Projects vs New Projects in Claude 6:03 Anatomy of Claude Projects: Under the hood 8:54 Claude Projects in Action (Demo Part 1) 16:12 How to use Claude Projects well 18:51 Boris Cherny and the Anthropic team's advice on Claude Projects 23:52 The Orchestrator Rule 27:23 Claude Projects in Action (Demo Part 2) 34:57 Mistakes to avoid (and managing usage limits)

Claude Code Projects let you describe what needs doing and spins up multiple threads in the cloud to do the work. Here's the pieces that make that possible: (1) GitHub repo: You need to connect a github repo for the work to happen in. (Pro-tip: you can work across multiple repos in a single Project) (2) Coordinator: "Chief of Staff" agent that splits requested work into threads, tracks it, and reports back. This is the big innovation of Claude Projects. You as the human no longer need to split up tasks or manage sessions. (3) Threads: Think of threads as worker agents. Each thread gets its own branch and its own cloud sandbox so they can make changes to the code without stepping on each others toes (like git worktrees). (4) Pull Requests: PRs are the unit of output for work done. PRs are made to your connected GitHub repo. (5) Shared context: The co-ordinator and threads all share the github repo, as well as instructions, memory, skills, and connectors. (6) Overview panel: Your control dashboard to see what's WIP, what's blocked on you and what's done. (7) Routines: threads that the coordinator can spin up to do work on a schedule. Full Claude projects anatomy walkthrough below:

Advice on how to use Claude Projects from the @ClaudeDevs team at Anthropic: @bcherny - "Projects have changed not only how I interact with Claude but how I code. I stopped managing sessions. I just send thoughts as they come." @_catwu - "[Projects] lets me operate at a higher level of abstraction. I fire off a bunch of tasks and move on. Claude has context on everything I'm working on and can give me an aggregated status update whenever I ask." @trq212 - "You can ask it to be proactive, to do things on a schedule...It feels a lot nicer than a bunch of sessions." @pranathiperii - "I'll usually toss 10-15 tasks to the project, let it cook, and come back to verify screenshots or steer threads as needed." @frankdotlee - "the coordinator form factor unlocked a totally different mode of working for me...one project per long-running workstream...pepper the coordinator with a bunch of tasks and then claude will work on them in parallel threads in the background." I cover them in more detail here:

What a day: Opus 5.5 GPT-6 Sol and Luna Cloudflare Workers Previews 🚀🚀🚀

Opus 5.5 is a very good model and I expect it will quickly become the daily driver for a lot of people. The writing style is so much better. It writes like Opus 4.6 and codes like Fable 5.1.

also important news we fixed the writing

I've always liked newsletters that let you read the latest edition first while signing up. So I asked Codex to add that to my weekly newsletter AI Native Builder Weekly. (No copy pasting involved, just told Codex to grab the latest edition from Kit via MCP)