Claude Chat is like a Desk

The desk is the context, and everything on it is something you chose to put there. Below is a finished one, set up and ready to send. Explore it as it stands, or walk through the five passes that built it, one object at a time.

The finished desk with every part labelled — instructions, projects, connectors, skills — all inside the context window.
Explain how it got there.
A bare dark desk — an empty mat, nothing written down — and Claude’s first draft stamped AI SLOP.

Bare desk

A clean context is like a clear desk

Confident, upbeat and unusable. With nothing written down, Claude picks its own voice, invents the numbers it lacks, and decorates the whole thing.

The desk with three instructions written on the mat; the reply is honest but stamped NO DATA.

+ Instructions

Instructions are like a desk mat — the rules and principles you always use as a foundation.

All three rules held — plain, no invented numbers, no emoji. The rules made it honest, not informed: it still knows nothing about Acme.

The desk with an open Acme project folder added; the reply is stamped STALE DATA.

+ Project

A project is like a folder of work — its own separate context, kept apart from everything else

Real numbers, real continuity, right frame — and a week stale.

The desk with connector apps added — live Drive and Slack screenshots — the reply stamped WRONG FORMAT.

+ Connectors

Connectors are like apps on the desk — the only thing here that can see today

Two facts that exist nowhere on the desk. This is the only lever that knows about today.

The full desk: instructions, project folder, connectors and an open handbook skill — a ready-to-send answer, no stamp.

+ Skills

A skill is like a page in the handbook — the way we do this exact job here

Same facts, your house shape. Nothing left to edit — that is the whole point of a skill.

What you're probably wondering

What is Claude?
Claude is the assistant. Anthropic makes it, I don’t, and everything this page walks through is already built into it, sitting in menus most people never open. There’s a free tier and a paid one, and you can have the bare desk from pass one in front of you in about two minutes at claude.ai .
What is WorkHoodie?
WorkHoodie is me, Sean. I’m the AI partner for companies that don’t have anyone in-house who can own this. Anthropic sells you Claude. I sell the decisions about what goes on the desk: which work is worth pointing at it, what the rules say, which connectors get turned on, and what finished looks like in your business. It starts with a two-week roadmap of where AI actually pays off, and I build and coach from there. Some businesses don’t need a desk at all, and if yours is one of them I’ll tell you that instead of selling you one. See how WorkHoodie works
How similar is this to ChatGPT or Gemini?
Close enough that you should check what you already pay for first. ChatGPT and Gemini both have projects, ways to connect your tools, and ways to save how a job gets done, so if your team already has ChatGPT seats, start there rather than buying a second thing to find out. What each one calls these and how far you can tailor them differs, so the desk looks a little different on each. The thinking carries over either way.
Can I make my own apps, plugins and skills?
Yes, and almost none of it is code. A skill is a page you write in plain English about how your company does one job. Plugins hand a stack of those pages to the whole team at once. The apps on the desk are Claude’s connectors, the links out to the tools you already run on, and the common ones are a switch you turn on, so when you need one nobody’s built yet, that’s the part I wire up. Your ops lead can write a skill in an afternoon, which also means someone has to update it when your process changes, and we settle up front whether that’s your team or me. Deciding which jobs deserve a page takes longer than writing the page, because a skill that describes the wrong process just makes the wrong thing faster.
How can Claude work for my whole team and not just me?
Everything on this page is one person’s desk. On a team plan you build one desk and everyone works from it: the same project folder, the same instructions, the same handbook of skills, instead of everyone running a private chat with its own rules. Seats are per person and you administer them like any other tool, so you can see who has access and cut it off when someone leaves. The real work is deciding what belongs on the shared desk and what stays on someone’s own.
Is our data safe once it can see Slack and Drive?
A connector signs in as the person who turned it on, so it reaches whatever that person reaches, and if that’s your ops director with access to everything, then it’s everything. The questions worth asking are who connects what and which folders that account can see, which is a governance call more than a technical one. Anthropic doesn’t train on business-plan data, and I’d have your ops or legal person read the policy themselves rather than take my word for it.
What happens when it makes something up and we send it to a client?
That’s the failure in pass one, and it’s the right thing to be scared of. Nothing on the desk makes it impossible. What the desk does is narrow it: a rule that says never invent a number, a folder holding the real ones, and connectors that go read today’s instead of guessing. Then a person still reads it before it goes out, the same way you’d read a junior’s draft. When we set yours up we decide which work is safe to send with a glance and which never leaves without a second pair of eyes, and that call is yours, not the tool’s.