The brief is the work

You have probably spent some time chatting with AI by now. You've worked out that it's pretty good for planning a trip, working out which Garmin to buy (just me?), or an alternative way to settle an argument than Google. For work it might have been less impressive, and more like, well, work. You start a new chat and start cold, explaining who you are, what you are working on, what "good" looks like, and what it ought to know. Then you close it, and tomorrow you do it all again.

That gap is familiar if you have ever brought in a consultant or an advisor (for some of you, that may be me!). They arrive with the craft but knowing nothing about you - your firm, your clients, how you like to work. The first conversation is not the work. It is onboarding: filling that gap so everything after it is useful.

AI is the same. Out of the box it gives you the generic answer - the polished, professional, ultimately useless version a consultant hands every client before they have learned anything about yours. The fix is the same too. You brief it.

A project is where that briefing lives, so you only do it once. Think of it as everything you would give a new consultant: a brief on how you work and what you expect, the reference materials they should draw on (your templates, your best examples, the standards you hold), as well as what they pick up about you as you go, so they start the next job further ahead than the last. Whether you are in Claude or ChatGPT* it is the same idea: a standing brief, so the AI stops starting from zero.

You can keep a project to yourself or share it, and when you share, the brief and the materials travel with it. That is how a business gets one standard instead of ten people briefing the AI ten different ways.

As with most things AI, the hard part isn't how to use the tool. It's being able to say what you actually want - what "good" looks like here, who the work is really for, and what you would never let out the door. Having both employed consultants and been one, I can confirm this is just as hard with people - the difference is that a person can fudge their way through (more work or a worse result is the usual price, but we will let that slide). AI gives you no such cover: it does what the brief says, so a vague brief shows up immediately. And the part that won't go into words - the judgement you can't quite explain - stays yours. Knowing where that line falls is its own kind of useful.

I found this setting up my own workspace, through more iterations than I'd like to admit. Setting up the project itself is instant; saying what I actually do - and, in my case, working out what I count as good - took far longer, because until then I had never had to put it into words. What changed once I had was concrete: instead of a polished, generic draft I had to redo, the AI started giving me first drafts that sounded like me and needed an edit, not a rebuild.

There is a sharper version if you carry key-person risk: the judgement that walks out the door when a senior person leaves is the most expensive thing you have never written down, and a shared project is the cheapest place to start capturing it. That is what compounds - not the tool getting cleverer, but your standard, written down once and reused by everyone.

The reassuring part is that you already know how to do this. You have onboarded people before. Where you start just depends on where you are.

If you are mostly new to using AI for work, do not start with something elaborate. Start with one general project - a single place the AI gets to know you and how you operate, the way a good assistant would. Give it a name, and let it interview you to build the brief:

I want to set up a general project to get the most out of [Claude / ChatGPT] for my work - one place where it gets to know me and how I operate. Before we start, interview me. Ask one question at a time about my role, my business, the work I do most often, how I like things done, and what I count as good. When we are finished, turn my answers into a set of project instructions I can save, the knowledge files needed, and suggest a short name for the project.

If you have been at this a while - a general project and a stack of chats behind you - the next step is to let the AI tell you when to specialise. Have it look across your work and recommend what to set up next:

Look across our recent chats and the work in this project. Group them by the kind of work I keep returning to, and tell me whether I should set up any additional, more focused projects. If so, name them, say what each should contain, and draft the starting instructions for the top one or two.

Whichever you are, the move this week is the same in spirit: have one conversation where you tell the AI how you actually work. The brief is the part that looks like admin and isn't. It is you, finally writing down what you can.

PS. If you’re using Google’s Gemini, there are no direct equivalents at time of writing. Gems have similar-ish functionality but are closer to saved personas than workspaces with dedicated memory.
If you’re using Microsoft’s Copilot, Notebooks are the closest match.