02 · THE OPERATOR'S EDGE · WEEKLY · BY MARC KLEINMANN

THE BREAKDOWN

📌 Your first Claude Project

Three things to set up before you use it for anything real.

The 30-minute setup most operators skip

Last week we walked the Claude map. Front porch, rooms, attic, basement, loading dock. The map answers what's where. This issue answers the question that comes next.

You've installed Claude. You have a Pro subscription. You've opened a chat or two. Now what?

The answer is not "use it more." The answer is: build a Project, put three things in it, and stop typing the same context into chat every morning.

A Project in Claude is a persistent workspace. You name it, you give it a standing brief, you drop your reference material into it. Every conversation inside that workspace starts with all of that already loaded. The chat window is the same chat window. The difference is that Claude already knows who you are, what your business does, and where to look.

The 30 minutes you spend setting one up right is the highest-leverage half-hour after install. Most operators skip it. Most operators are also the ones still pasting their company description into every new chat.

Three things to set up:

1. The Project · a named workspace

2. The Instructions · a standing brief

3. The Knowledge · the filing cabinet

1. The Project itself

Open Claude. Sidebar. New project. Name it for the work, not the tool. "Praxis - client work" beats "Claude project." "Acme Dental - operations" beats "AI assistant."

The name is not cosmetic. You will have more than one Project within a month. Naming for the work is what lets you find them later. I have nine right now. The ones I named generically I have to open to remember what they are. The ones I named for the work I can scan.

Pick one purpose per Project, not one per tool. Marketing, operations, hiring, finance. Those are Projects. "ChatGPT replacement" is not a Project. The framing matters because the Instructions and Knowledge you put inside only earn their keep when they're scoped to a real job.

2. The Instructions

This is the load-bearing piece. Most operators leave it blank. Then they wonder why Claude doesn't sound like them, doesn't know their business, treats every conversation like it's meeting them for the first time.

The Instructions field is a standing brief Claude reads at the start of every conversation in the Project. It's the thing the front porch is missing. It's the difference between "stranger who's read a lot" and "colleague who's been here three months."

Four things go in it:

1. Who you are. One paragraph. Role, business, what you sell, who you sell to.

2. How you talk. Direct or formal. Long or short. Bullet-heavy or prose. Em dashes or no em dashes. Voice rules carry across every conversation in the Project.

3. What you're working on right now. Current priorities, current clients, current campaigns. Update this monthly.

4. What to avoid. Phrases that sound off-brand. Topics that aren't yours. Decisions that aren't Claude's to make.

Here's one filled in. Steal the shape, change the specifics.

Length target: 200 to 600 words. Long enough to be useful, short enough to read. If you're staring at a blank box, start by pasting your About page from your website. Edit from there.

The tell that yours is working: the third conversation in the Project feels like the third conversation. Not the first one again.

In your other tool. ChatGPT has Custom GPTs and the project-level custom instructions. Gemini has Gems. Same primitive, slightly different shape. Saved instructions per workspace.

3. The Knowledge

The third thing is what Claude searches automatically. Upload your reference material here: the standing documents your work pulls from. Brand guidelines. Pricing sheets. Service descriptions. Client templates. The five PDFs you keep emailing yourself.

This is not a dump. Three rules.

1. Curated, not exhaustive. Twenty good files beat two hundred mediocre ones. Claude searches everything in the Knowledge folder; noise dilutes the signal.

2. Refresh on a cadence. Pricing sheets go stale. Brand guidelines get rewritten. Set a monthly calendar reminder to audit what's in there.

3. Operational documents, not aspirational ones. Your actual rate card, not the one from three years ago. Your actual SOP, not the one you keep meaning to write.

Pro tip: pre-curate before upload. Strip a 200-page PDF to the 12 pages that matter. Save spreadsheets as cleaned CSVs. Don't dump a folder of zips and screenshots and hope. The cleaner the knowledge, the better the answer.

The tell that yours is working: you ask a question, and Claude often references the specific document. "Per your service description v3, the discovery call is 30 minutes." That citation is the system earning its keep.

The operator-altitude take

A Project is not a productivity hack. It's a context shift.

The front-porch version of Claude is one-off conversations. Smart but unanchored. Every chat starts cold. You're the connective tissue.

The Project version is anchored. The same Claude, the same chat box. But every conversation starts with your business loaded. The connective tissue is the system.

The thirty minutes is not the cost. It's the threshold. The cost of NOT setting one up is paid in compound interest, one stranger-conversation at a time, for as long as you keep using the tool. Most operators pay that bill for months before they notice.

The deeper version of this issue lives at praxisx.co/edge/get-started-with-claude. The install playbook. Same argument, one level deeper.

SIGNAL

🔍 Signal

Three reads worth your time this week.

SIGNAL 01 · ANTHROPIC · APRIL 28

Claude for Creative Work

Anthropic shipped connectors into creative tools like Blender, Adobe Creative Cloud, Autodesk Fusion, and Ableton. Same Claude, same chat box, with the tools pre-wired and workflows pre-loaded for one vertical. That's the macro version of what you just did at the desk-scale this morning. A Project is the personal version of a vertical pack. You name the work, you load the brief, you load the reference shelf. Anthropic is doing it for creative pros at the platform level. You're doing it for your business at the workspace level. Same primitive, different floor.

SIGNAL 02 · TECHCRUNCH · MAY 13

Anthropic courts a new kind of customer: small business owners

Mainstream press named the framing trade press has been missing. AI tools are rarely tailored to how small businesses actually operate. Most of what you read on the AI shelf is enterprise sales copy lightly retouched for SMB. The Project you set up this week is the inverse move. You tailored Claude to how YOUR business operates. Pricing sheet, brand rules, the actual SOP, the four bullets on what to avoid. That's not what an enterprise pilot looks like. That's what an SMB Project looks like.

SIGNAL 03 · PETER YANG · LINKEDIN · MAY 8

The chat era is coming to an end

Yang names the shift. Anything bought today as "AI chatbot" is depreciating fast. The replacement is an agent that gets you. You just set up a Project. That Project is your first agentic flow, before you knew that's what you were building. The Instructions field is the standing brief. The Knowledge folder is the reference shelf. The chat window is the surface. Same primitive, just renamed. Build the flow now while the rest of the market is still buying the chatbot.

TRANSLATION

✏️ Translation

Operator voice from outside the AI shelf.

A line worth carrying, translated into what it means for an SMB operator.

"I run three businesses. AI isn't running them for me, but it's become a genuinely useful tool in the rotation." · Zac Wolf · LinkedIn · May 15 (11 days old) · 23 reactions.

Zac Wolf runs three small businesses. A photo booth company. A vintage matchbook print shop expanding into wholesale. A wedding photography practice. He wrote a quiet post about how AI shows up in the rotation, and the line that landed was this:

Pocket Fire Print Co. needed a real inventory system. I have a collection of vintage matchbook prints and zero desire to manually catalog them. So I connected Claude to my image folder and had it build out a full spreadsheet: inventory numbers, names, categories, image links, the whole thing. Days of work, handled.

That is the Issue 02 argument in operator language.

The thirty minutes you spend setting up a Project is the same trade Wolf made on the inventory build. Took him longer than doing it manually once. Now it's done forever. Stores in Orlando, NJ, CA, and TN can place wholesale orders against a real backend instead of a notebook.

The operators who set the thing up once buy themselves compounding time. The ones who keep typing context into chat every morning pay the bill in single-conversation increments, forever.

Wolf's word for the trade was the right one. Every time.

Read it: Zac Wolf on LinkedIn →

SIGN-OFF

✉️ Talk Wednesday

See you next week.

Two questions for the reply box: Did you set up a Project this week? What went into Instructions?

The setup walkthrough lives at /edge. If you want to take this from concept to operational, the playbook is at praxisx.co/edge/get-started-with-claude. Installing Claude, mounting Cowork, wiring the connectors, the about-me file that makes any of this work. Same argument as this issue, one level deeper.

Talk Wednesday.

Marc