Context-Engineering a Company Agent
Why every company needs to be written down for AI, and why describing a company is so hard.
Everything has to be written down
In the new AI era, it is essential that all of a company's processes, rules and intentions are written down. That takes different file formats, so that everything is structured for LLMs.
In the world of programming, there are already many standards for this:
AGENTS.md: a file in a repository that tells coding agents how a project works and which rules to follow.- Skills: packaged instructions and resources an agent loads when a task needs them.
- Plugins: bundles that extend an agent with new commands, tools and behaviors.
- MCP (Model Context Protocol): an open standard for connecting agents to external tools and data.
Understanding a company down to the last detail
For every CEO, it is important to understand their company down to the deepest detail. Based on that understanding, they can make decisions or pass that knowledge on.
But companies and company cultures are usually very complex, and truly describing a company is very hard. I think it's comparable to describing a person down to the deepest detail: their character, how they think, how they act.
The question
So how do you do that for a company, so that an agent can use it as context and actually help the company in a meaningful way?
How can you use context engineering to build a company agent?
The original notes
This is how the idea started, as notes in a notebook. Click a page to enlarge it.