Your AI forgets everything about you. Here’s how to give it a memory.

Generative AI is brilliant at answering the single question. In real work, though, it trips over something basic: continuity. What you decided in March has to count in June, and a conversation that restarts from zero every time doesn’t give you that. You end up re-explaining, every time, who you are, what you sell, what you’d agreed on.

Meanwhile your company’s context stays scattered: some in your head, some in ten different chats, some in folders no one can find. It’s not a detail. According to McKinsey research, a person spends nearly a full workday a week just searching for internal information[1], and on average we switch applications more than a thousand times a day[2]. Every time you rebuild the context for the assistant, you’re paying that same tax.

The answer isn’t a smarter AI. It’s giving it a place to keep the memory. Instead of trusting the context to your head, you put it in tidy files the assistant reads before it answers and updates afterward.

Everything has its place: who you are and what you do, how you position yourself, the decisions made, the questions still open, the work in progress. The AI doesn’t remember by magic: it rereads the archive. The tidier you keep it, the more precise it gets, and the less time you waste repeating yourself. It’s the same logic as the single source of truth you’d use for a team, one, instead of knowledge scattered across heads, chats, and spreadsheets.

A tidy archive ages fast if no one updates it. That’s why the system stands on a single rule, and it’s worth learning even if you never use any tool: at the end of every important task, the assistant checks whether something worth remembering has come up, a decision, an open question, a risk, a task, and writes it in the right file.

It’s tedious. It doesn’t make for a great demo. But it’s exactly what separates a workspace that gets more useful every week from a chat that always starts over. Without this habit you’re back to a blank page; with it, every Monday the AI knows a little more about your company.

That’s the method. Let’s see it with the tool we actually use, Claude Cowork. You open a project and give it access to a folder on your computer. Then you don’t start by filling in forms: you upload what you have, even if it’s a mess. Old notes, a quote, a client’s email, a few scattered files.

You tell it “organize.” Cowork reads everything, separates the facts from the assumptions, and files each thing in the right place: the company profile, the offering, the tone of voice, the decisions. At the end it leaves you an honest summary of what it understood and, above all, of what’s missing. From that point on, when you ask it for an analysis or a piece of content, it starts with your context instead of a blank page.

Once the workspace has a memory, some things you stop doing by hand. Cowork can run them on a schedule and leave you a ready summary: the marketing review on Monday morning, a brief with the agenda and the things to handle, a site check on Friday.

You read and decide; the gathering is done by the assistant. One common-sense rule we always keep: anything that goes out to the outside world, an email or a message, stays a draft. Nothing ships without your go-ahead.

This method organizes the work and makes it repeatable, but it doesn’t solve everything, and that has to be said. When a process has to run in production, at high volume, connected to your back-office systems and without a person hitting “send” every time, you don’t need a tidy chat: you need real automation, built and put in place.

That’s the moment you move from DIY to custom software. Knowing it in advance saves you from forcing the wrong tool onto a bigger problem. If you’re at that point, it’s exactly the work we do: an automation that runs on your data, with code you own and fixed timelines.

You don’t have to build everything from scratch. We packaged our system and made it open source. It’s called cowork-os: the folder structure, the instructions, the automations, and a real, already filled-in example.

You copy it into your Cowork project, paste in a file, answer six questions, and Claude builds the workspace to fit you. Even if you’ve never written a line of code. And if you’re starting from a messy folder, you have it tidied up in a single pass. You’ll find it free on GitHub.