Explainer · FOUNDATIONS / INTRO
How agentic work works
The whole mental model in one place: the three capabilities, what an agentic workspace actually is, and the building blocks you will hear about, skills, CLIs, and MCP.
When you work with an agent, you are not using a chatbot. You are running a workspace: your context, your tools, and an agent that can act on both, with you in control. Three ideas make sense of all of it.
Three capabilities, not fifty tools
Most people feel behind because they chase tools. The people who move fastest understand three capabilities instead. Click each one.
Context
Organize your company knowledge so the AI has something real to work with: documents, decisions, processes, history. Without it, even the best model is just guessing.
for example Your proposals, past projects, and brand voice in one place, so a draft sounds like you on the first try.
The tools change every week. These three do not. Everything else is detail.
What an agentic workspace actually is
Not a chat window. A workspace puts your context at the center and lets an agent act on it. You ask, review, and approve; the agent reads, acts, and reports back.
Two things make it different from a chatbot:
- It remembers, sees your files, and can act. A chat window does none of that across sessions.
- It is yours. Your repo, your context, versioned and inspectable. Not locked inside a vendor you cannot leave.
The building blocks you will hear about
Three words come up constantly once you start. Here is what they mean, in plain terms. Click each one.
Skill
A packaged capability the agent can reuse: instructions plus the code and tests to do one job the same reliable way every time. When something works once, you 'skillify' it so it never has to be figured out again.
for example A "send the weekly update" skill that always pulls the right numbers, formats them the house way, and drafts the email.