At some point over the last few months, Anthropic began curating its ecosystem, releasing updates at an incredible pace. As of last week, the number was 74 releases in 52 days, and the drops haven’t stopped. This is not a complaint. It is, if anything, the opposite of a complaint. But I understand why people who open Claude for the first time right now just want to know which button to push. Where exactly do you start?

After spending more time inside this ecosystem than is probably healthy, I can tell you: you don’t start with the features. You start with the map logic. The underlying structure of how Claude is designed to work hasn’t changed, even as the surface keeps expanding. Once you have that structure, everything Anthropic ships drops into place instead of adding to the confusion.

The ecosystem has three layers. Claude Chat and Projects at the base. Cowork in the middle. Claude Code at the top. Each one extends the same underlying logic further into your actual work environment. Each becomes obvious once you’ve built the layer below it, because the mental model is identical at every level, just with more capability behind it. If you skip to layer two without building layer one, and you’ll get output that works fine and feels wrong, and you’ll just blame the prompts.

Importing your Memory

If you're new to Claude and you want it to know you, go to Settings > Capabilities > Memory, and Claude will give you a tool to import Memory from another LLM. You can edit and add to it. It will tell Claude who you are before any conversation starts: your role, your communication style, what you're working on, what you care about. Think of it as the cover letter you write before handing someone your portfolio. Claude now arrives at every conversation with a baseline understanding of you rather than starting from zero. The difference in output quality from that one setup step is immediate.

Projects

Layer one is a Project, and the most important thing to know is that it has two components that do completely different jobs. Project Instructions are behavioral: how Claude communicates, what it optimizes for, and what it never does. Project Files are contextual: what Claude reads before responding. Brand voice, audience profiles, past work, research, and briefs. Instructions are rules. Files are memory.

Before adding anything to a Project, I write the instructions: the role I’d like the LLM to assume, what success looks like for this specific project, and what is off limits. The files come next, and Claude can help you create the files that it will read later. The system helps you build the system. Once you see that loop operating, the whole process accelerates.

Cowork

Cowork is layer two, and it uses the same folder structure as your Project, but it's on your actual machine. Claude reads your working files where they live. Output lands in your file system instead of a chat window. The work connects to your real environment instead of evaporating into a thread you’ll never find again. What this unlocks is the context problem, the thing that causes generic output, gets solved with better files. A well-organized Cowork folder does what longer prompts never quite manage:  the output gets more specific, and the prompts get shorter.

The folder logic was what made me fall in love with Anthropic. I'm a zero-inbox person. I have a folder for my folders. The idea that building a good Claude system is really just building a good file structure scratched something deeply specific in my brain. If you are also the kind of person who reorganizes your desktop before starting a new project, this ecosystem was designed for you.

Before Cowork generates anything on a complex task, ask Claude to ask you questions first. For strategy documents, creative briefs, anything where the right output depends on assumptions you haven’t articulated out loud, that clarification step surfaces the gaps in your own thinking before they show up as vague output. Remember, it may be annoying and take some time, but the time invested upfront will be paid back exponentially. Have a look at the Anthropic plug-ins in Cowork created specifically for knowledge work. These will give you a leg up and boost your system out of the gate. (I am particularly fond of the productivity plug-in, if you couldn’t guess.)

One of the most important files you can create for the ecosystem is your About Me file (sometimes called claude.md or soul.md). I spent an afternoon in a therapy session with Claude, talking about my goals, beliefs, values, and business, to set this up. It was quite cathartic, but it also helped Claude understand how I really work and how to push me and itself to get what I am looking for.

Claude Code

Layer three is Claude Code, and I want to be honest: most people reading this don’t need it yet. Cowork covers most creative operations and knowledge work. Claude Code is for when you want the output to be built rather than written, or when you want multiple agents working across the same project in parallel. What sounds technical about it mostly lives in the description. The actual practice is simpler: you write text files that describe jobs, put them in the same folder you’ve already organized, and Claude Code orchestrates across them. A research agent, a fact checker agent, a writer agent, and one formatting output. Each one is a markdown file. The mental model is the same one you built in layers one and two, just with more autonomy.

I put off touching Claude Code for three months because the word "terminal" made me feel like I was about to break something. But with the introduction of Code into the Desktop app, I realized I just had to tell Claude what I wanted to do, and it would walk me through it.

Skills

Underneath the three layers is a concept called Skills. This is the part of the ecosystem that actually makes the work feel portable, though Anthropic's naming of it is a bit dry. Think of a Skill as the "final form" of a prompt you've finally stopped fighting with. You don't set out to "design" one; that's a trap. You just wake up one Tuesday and realize the prompt you've been using to generate weekly briefs hasn't changed in six weeks. It's done. At that point, you ask Claude to package it as a Skill. After that, it just loads.

I've found that every useful Skill I own started as a messy, 500-word prompt I kept aggressively rewriting until the output stopped feeling like a roll of the dice. If you're looking at your own workflow, don't ask "What could be a Skill?" Ask: "What am I sick of copy-pasting every Monday morning?" That's your Skill right there.

As I’ve mentioned before, Anthropic and many other creators have public resources for skills. So one of the easiest ways to create a skill is to find one that is close to what you are looking for, duplicate it, and edit it for your use case.

Model

Extended Thinking is the setting most people skip. Turn it on for strategy, reasoning, anything where the quality of the thinking is the deliverable. I am very precious with my Claude limits (I hate the dreaded “you’ve timed out” message), so I use Sonnet for most stuff, but I break out Opus when I’m actually confused.

The Reality Check

Anthropic is going to keep shipping updates at a pace that is actually a full-time job to track. It's noisy, and it's fine to feel a little behind. Get your files in order. That's the goal. If your context is a mess, the smartest model in the world is just going to give you really confident, well-formatted garbage.

Start with ten minutes. Don't try to build the "ultimate" system today, just pick one project that’s currently sitting on your desk and move it into a folder. Once that clicks, the rest of the ecosystem starts to feel less like a technical shift and more like a better way to work.

This stuff moves too fast to do alone. If you’re sitting there looking at a blank Project window and feeling like you’re doing it wrong, you aren't. Send me a screenshot or a quick description of what you’re trying to build, and I’m happy to take a look and tell you where I’d start.

— Lauren Eve Cantor

thanks for reading!

if someone sent this to you or you haven’t done so yet, please sign up so you never miss an issue.

I’ve also started publishing more frequently on LinkedIn, and you can follow me here

if you’d like to chat further about opportunities or interest in AI, or this newsletter, please feel free to reply.

banner images created with Midjourney.