Execution · Proving to Scaling
A company brain
Your own AI fetches the state of your business from one source, with your permissions, so you stop hunting for it or pasting it in.
Worth reading first: A project with your own documents
Does this sound familiar?
- Before the weekly meeting you copy the state of your priorities from three screens into ChatGPT, and next week you do it all again.
- A member of your leadership team asks how a client project is going, and you have to ring the project manager before you can answer.
- You ask AI for advice on your quarter, and the answer is about a company it doesn't know.
- You're not sure whether you're allowed to paste your figures into an AI app, and then you do it anyway.
Who it's for, and when
For founders, members of the leadership team and managers who already use Claude or ChatGPT and keep pasting the same figures into it. It helps once there are priorities, KPIs and tasks that people keep up to date, so from Proving onwards. In Searching there's not much to write down yet; the business still lives in your head, and that's fine at that stage.
The model
A company brain is what your business knows, in one place your AI can read: your strategy, your priorities, your KPIs, your tasks and your ways of working. Your AI fetches it itself when you ask something. You don't paste anything any more, and the answer is about where things stand today.
It only works with permissions. Your AI sees what you're allowed to see, and no more. If an employee asks about a colleague's performance review, they don't get it, just as in the app. That way everyone in the business can connect their own AI without anyone seeing something that isn't meant for them.
In Boldcaster this is the AI connection. You add Boldcaster to Claude or ChatGPT as a connector and log in with your own account. The owner or an admin switches the connection on for the organisation, chooses which parts the AI can see, and whether it's read only. Writing always takes two steps: the AI shows a proposal with what's there now and what it'll become, and it only goes ahead after you say yes. Every request is in the log, including the ones that were denied.
The lesson A project with your own documents is about documents that don't change often, like a quote template or your brand guidelines. Anything that changes every week doesn't belong there, because an upload is out of date after a week. You read that through the connection. How to ask the question itself is in the lesson Asking AI a good question.
The problem is older than AI: Ikujiro Nonaka and Hirotaka Takeuchi, The Knowledge-Creating Company (1995), on knowledge that lives in people's heads and knowledge that's written down in one place. AI has been able to read from that place since November 2024, through the Model Context Protocol (MCP), an open standard from Anthropic that ChatGPT supports too.
1. One source
What's been agreed is in one place. If the state of a priority is also in a spreadsheet and in someone's email, your AI doesn't know which one is right, and neither do you.
2. The same permissions
Your AI sees what you see. The owner or an admin decides for each part what the connection makes available, and for each connection you choose which workspaces it sees.
3. Writing in two steps
Your AI can propose a task or a KPI figure. It shows you what will change, you say yes, and only then is it saved. If the connection is set to read only, it writes nothing.
4. Keeping it up to date is the owner's job
The AI reads what people keep up to date. If nobody updates their priority, you'll get an out-of-date answer that sounds very sure of itself. The current state stays the job of whoever owns the priority or the KPI.
Pitfalls
Letting the AI decide
You ask which priorities are behind, and the answer is useful. Deciding what stops or what gets added is still a person's job. Let your AI prepare the conversation, and have the conversation with the owner yourself.
Thinking nothing leaves the business through the connection
Whatever your AI fetches through the connection ends up in the conversation with the AI provider, just like pasted text. Anthropic and OpenAI, the makers of Claude and ChatGPT, are American companies. The difference from pasting: the AI only gets what you're allowed to see, and the log records what it asked for. So don't make a part available that you don't want to end up there.
Watch and read
How to add a connector in Claude and what the AI can do with it afterwards: exactly the step you take for Boldcaster too.
Watch on YouTubeWhat the Model Context Protocol is, in four minutes: a standard way for an AI app to request data from another system.
Watch on YouTubeThe same idea in ChatGPT: answers from your own company's systems, with the source alongside.
Watch on YouTubeRead
The Knowledge-Creating Company
Ikujiro Nonaka en Hirotaka Takeuchi · 1995 · ISBN 9780195092691
The classic on knowledge in a company: what lives in people's heads and how you make it shareable. Written long before AI, which makes it a good reminder that the problem is older than the technology. 284 pages.
View the bookApply it
- 1
Have the owner or an admin switch the connection on
Under Integrations, on the AI connection tab. They'll agree to the AI connection terms there first. Start with read only and with the parts you really need, such as priorities, KPIs and weekly tasks. You can always add more.
- 2
Connect your own AI
Add Boldcaster to Claude or ChatGPT as a connector using the address on the screen, and log in with your own account. The same screen explains how it works for each app; in ChatGPT it depends on your plan. If you have more than one workspace, choose which ones this connection sees.
- 3
Ask one question you now answer by hand
Take the question you answer again every week, like where things stand before the weekly meeting. Check the answer against the screen in Boldcaster. If it matches, you've got your first regular question.
- 4
Check the log
It shows what your AI asked for and what was denied. A denial usually means a part hasn't been made available, or your role isn't allowed to see it.
Examples
Questions to ask your own AI about your business
Which priorities are behind this quarter, who owns them, and which tasks under them are still open? Which KPIs are off track or need attention, and since which week? What's on my team's board this week, and what's past its deadline? Get the state of things ready for Monday's weekly meeting: one line per priority. What's our way of working for a new client, and which version is it? Where did we write down that agreement about the handover? Search the tasks and the comments. Which of my own tasks won't I get done this week? For every answer, ask where it came from. If a number doesn't match the screen, the screen is right.
Wrong and right side by side
Wrong: you copy the KPI scoreboard into a chat and ask 'what stands out?'. Next week the copy is out of date, and nobody remembers what you put in it. Right: 'Fetch this quarter's KPI scoreboard from Boldcaster and list the KPIs that are behind, with their owner.' Your AI requests it itself, with your permissions, and the log shows what it asked for.
Creating a task from the conversation
You say: 'Create a task for Sanne: quote for the new client, deadline Friday.' Your AI shows the proposal: title, owner, deadline, and which workspace. You say yes, and the task is in Boldcaster. If you say nothing, nothing happens.
Do this in your own business
Everything up to here is free. What you fill in for your own business, you keep in Boldcaster.
- Go to the AI connection. This is where the owner or an admin switches the connection on and chooses what the AI can see. You'll find the address for your connector, your connected apps so you can disconnect them, and the log.
- You look back on a situation from your own work. Only you can read what you write.
- You set a date with yourself, and on that day the app asks how it went.
- You assign the lesson to your team and see who has done it.
- The quiz counts towards your progress, and you see which lesson comes next.
Four weeks free. You'll land straight in this lesson.
Quiz
6 situations. You need 5 of 6 right. You can try again at any time.
1.You want ChatGPT to help you prepare the weekly meeting. What works best?
2.An employee connects their own Claude and asks about a colleague's performance review. What happens?
3.You tell Claude: 'mark the priority New website as completed.' What happens first?
4.Your AI says a priority is on track. In the meeting it turns out it's been stuck for three weeks. Where's the problem?
5.You're switching the AI connection on for your organisation for the first time. Where do you start?
6.Your AI says three priorities are behind and suggests dropping one. What do you do?
Nothing is saved here. In the app, the quiz counts towards your progress.
Read next
- Skills for recurring work · You set up one job you do often as a skill, so your AI does it the same way every time without you explaining it again.
Not sure where to start? Take the growth scan