Architecture & Governance
We build the data structure your company runs on, so AI can navigate it the right way: what lives where, who can do what, and what the AI may touch.
The data traffic light
The first thing we write is one page with three colors. Green is public or generic information, free to use with any AI. Yellow is internal working information, allowed only inside the company's official tool, in the company's own account. Red never leaves your internal systems: personal data, payroll, financials, recipes and formulas, contracts. In this phase, none of it enters an AI tool. Every employee reads the page once and knows in five seconds what they can use with AI and where.
- The red list is yours: we bring the rule, you decide what goes in it
- It's the first rule written, and every other rule hangs from it
ICM: the structure AI can navigate
ICM (Interpretable Context Methodology) puts the company into folders an AI can walk: context, processes, knowledge, clients, outputs, each in its place, each with one job. The structure decides what the agent reads; the agent doesn't. Every fact lives in one file and everything else points to it, so the answer is the same whoever asks. A fresh model opens the folder, follows the index, reads what the question sends it to read, and answers citing the file it found.
- Plain text, readable by any model and by any person
- Portable: the folder is a zip. If a provider goes down, you switch and nothing changes
- Every output is a file a person can open and correct before the next step reads it
- If the AI disappeared tomorrow, a process people can read is still there
Three levels, one structure
Each person has their agent. Each area has its brain. Management has the super agent. The three sit on the same structure and the same rules, so what a person's agent does shows up in the area's brain, and what every area knows feeds the super agent. It answers with information from all the areas, and each new agent added below makes it smarter. The order is one of dependency: without people who know how to work with their own agent, there's nobody to feed the levels above.
- Their agent: does the repetitive work of one post. The person reviews and decides
- Its brain: the area's agents, KPIs and approvals, visible to whoever leads it
- The super agent: every area, every agent, every KPI, cross-referenced
Governance: who can do what
Roles and permissions get written before anything is built. Who can read what, who can write, what needs a person's sign-off, and what the AI may touch at all. The matrix is short on purpose: three levels, a handful of actions, and a red column that stays empty. For companies with several sites, the map also says what does not travel between sites, in writing, so a rule that works in one place doesn't leak into another by default.
- The layer map: what exists, what lives where, what connects to what
- The governance model: who can do what, and what happens with the sensitive
- What does not travel between sites, said explicitly
- The material for you to defend it internally without us
What it connects to
The structure sits on top of the tools you already run: email, Drive, spreadsheets, the CRM, the ERP. It reads from them and writes back where the rules allow. Every account is in your name from day one. We create and configure them, and we can run them during the start, but we host nothing for you and the costs are yours. The day you want to take over, there's nothing to migrate.
Phases, not dates.
The order is fixed. The calendar is set once the diagnosis closes.
- Phase 01
Map
One hour with each area lead. We list the systems, where each kind of data lives today, and who's already using AI on their own.
- Phase 02
Write the rules
The traffic light, the roles and permissions, and what doesn't travel between sites. On paper, before anything is built. You sign it off.
- Phase 03
Build the structure
The folders, the one-source-per-fact library, and the connections to your systems, all in your accounts.
- Phase 04
Hand over
The material to defend it internally without us, and the sequence of what gets built next, in order. The calendar is set once Phase 01 closes.
Little, and early.
A short list, all before the first session. None of it takes a day.
- 01Your org chart: areas, who leads each one, and the sites if there's more than one
- 02The list of systems in use, including the AI tools people already use on their own
- 03One hour with each area lead
- 04Your data policy, if there is one. If there isn't, the traffic light becomes the first version
- 05One person who can decide the rules. If there's a committee, the name of who speaks for it
- 06Admin access to create the accounts in your name
What people ask about this.
Usually not. The structure sits on top of what you already run: email, Drive, spreadsheets, CRM, ERP. What changes is where each thing lives and who may touch it. If a tool is the reason data leaks, a shared spreadsheet with payroll in it, say, we'll tell you, and the fix goes into the plan.
You do. It's folders and plain-text files in your accounts, readable by any model and by any person. If you change AI provider, the folder stays the same. If we stop working together, there's nothing to migrate.
It goes in the red column and stays in your internal systems: personal data, payroll, financials, recipes and formulas, contracts. In this phase none of it enters an AI tool, and the permissions matrix says so in writing before anything gets built. The list is yours to extend.
The second brain is one executive's assistant, a quick win built on the context that person already has. Architecture & Governance is the structure underneath: the areas, the rules, the super agent that answers with information from every area. You can start with either. If you start with the brain and the company grows into it, this is the next step.
Ready to put order under your AI?
One discovery call. We tell you whether you need the full map or just the traffic light this quarter, what it takes and what it's worth. If the answer is no, we say so.