The four building blocks of a company brain

Ask a new colleague to “make the monthly marketing report” and you will spend the first hour explaining. Which template. Which channels. Where the numbers live. How the brand should look. An AI model starts in the same place: capable, but it does not know how your company works until someone tells it.

A company brain closes that gap. Zaela, the company brain built by Edge, splits it into four building blocks: memory, agents, skills and tools. The split is not arbitrary. In its guide to building effective agents, Anthropic describes the basic building block of agentic systems as a language model “enhanced with augmentations such as retrieval, tools, and memory”. Each block answers a different question.

Memory: what it knows

Memory is shared knowledge, organised per topic or brand: brand rules, tone of voice, colours, logos, guidelines, important links and earlier work. In Zaela it keeps a version history, so nothing disappears quietly.

For the monthly report, memory supplies two things: the report template and the brand rules. That is all the report needs, and that matters. Anthropic’s engineers describe a model’s context as “a finite resource with diminishing marginal returns”: more text is not automatically better. Knowledge kept per topic lets the AI bring in what this request needs and leave the rest.

Agents: who does it

An agent is a role, like a colleague who is already trained: a content writer, a report builder, an SEO specialist. Agents know each other and pass work along.

The monthly report goes to the report builder. Another request can need two roles: “every Monday, mail me if our rankings drop” brings in both the schedule planner and the ranking monitor.

Skills: how it’s done

A skill is a method, written down once and used by everyone: how to write like a human, how to translate meaning instead of words, how to make a reel that stops the scroll. An agent is the who, a skill is the how, and several agents can share the same one.

Anthropic, which uses the same word for its own Agent Skills, puts it well: building a skill for an agent “is like putting together an onboarding guide for a new hire”.

Not every request needs one. The monthly report runs on its template and its role. A LinkedIn post in your tone does use a skill: the content writer applies human writing, so the post sounds like you.

Tools: the hands

Tools are the connections to your platforms and data, through one gateway. Your people connect one link, and the right tools come along.

For the monthly report, the tools are Google Analytics 4, Google Ads, Meta Ads, LinkedIn Ads, Search Console and a calculator. The calculator is there for a reason: every ratio is recalculated from the raw numbers, with exact sums instead of mental arithmetic.

All of this works inside the AI your team already uses, through MCP: in the words of its own documentation, “an open-source standard for connecting AI applications to external systems”.

How the four combine

Follow the monthly report from question to result:

1. You ask, in plain language: “Make our monthly marketing report.”
2. Your AI, Claude or ChatGPT for example, conducts. It links the right knowledge, role, method and tools.
3. Memory supplies the report template and the brand rules.
4. The report builder takes on the work.
5. The tools read every channel and recalculate every ratio.
6. You get an Excel report with every channel, every ratio recalculated from the raw numbers, and a short summary of what worked.

Change the request and the mix changes. Turning your brand film into a 30-second reel uses the brand rules and the brand film from memory, the video editor as agent, the reel maker as skill, and video analysis, image and video generation and Gemini as tools.

A simple test tells you where something belongs:

• Something your team knows goes into memory.
• Someone who owns a kind of work becomes an agent.
• A way of doing things becomes a skill.
• Access to data or an action becomes a tool.

Why access per person matters

Once an AI can read your ad accounts, your CRM and your files, the question is no longer only what it can do. It is also: for whom?

Zaela answers that per person. Your AI sees exactly what you may see, nothing more. Keys are encrypted and never visible, not even to the AI. Every action is logged: who did what, through the website or the AI. Costs stay under control, with budgets and limits per person.

This is the security principle of least privilege, which NIST describes as restricting the access privileges of users “to the minimum necessary to accomplish assigned tasks”.

It is also what makes a shared brain possible at all. Without it, you would have to choose between a brain that knows too little to be useful and one that shows everyone everything. With it, the whole team shares the same agents and methods, while each person reaches only the data they are allowed to see.

Where to start

In Zaela, everything the AI uses, you can also see, adjust and add to in the web app. Take one request your team keeps repeating, and look at which of the four blocks it needs. See how a request flows through Zaela.