Your AI ecosystem, not a store
The mistake is thinking about all of this as "a place to put our documents." What actually compounds over time is not a store, it is an ecosystem: several layers working together, each one pulling its weight.
The whole picture
Here is the single picture of how it all fits together, top to bottom.
Static we curate and gate. Dynamic we leave at its source and read it live. Skills pull from both, through the connectors, wherever people are working.
At the top, people engage with the whole system through the tools they already use. Below that, skills do the actual work, one real job each. Underneath, connectors are the one way everything reaches into your real systems. And at the base, your knowledge sits in the two kinds you already know from the last chapter: static and dynamic. Everything above draws on that foundation.
Where people work
Claude is our default, the place most conversations and requests happen. But it sits alongside the other tools your team already uses day to day, whether that is a code editor, a messaging app, or anything else people reach for out of habit. The point is not to force everyone onto one surface, it is to make sure the layers underneath are reachable from wherever people actually are.
The doers
Skills are the layer that does one real job each: drafting a proposal, prepping a meeting, pulling together a weekly report. Kept narrow and well-defined, a skill is easy to trust and easy to reuse. The next chapter goes deep on how skills stack on top of each other.
One way in
Connectors are how the layers above reach your systems. The way we've found it useful to organise this is a single custom MCP server we own as the one gateway into our internal systems, with third-party vendor MCPs called directly for everything else. That split is our own operating choice, born out of wanting one guarded doorway into the systems that matter most, not something Claude itself prescribes.
The foundation
Underneath it all sits knowledge, in the two kinds from the previous chapter: static knowledge, curated and gated in one small, trusted place, and dynamic knowledge, left at its source and read live rather than copied. Get this foundation right and everything built on top of it, skills, connectors, the tools people work in, gets better with it.