Guides — interactive, grounded in the catalog
Learn the landscape
The agent market is easier to read once you have the right dials. These explainers teach the two that matter most — how autonomous a system really is, and which capabilities your task actually requires — using the live catalog as the example set, not hypotheticals.
Interactive guides
The Autonomy Dial
Autonomy isn't binary — it's a dial. Slide from 'suggests text' to 'holds a job' and watch what changes: who initiates, where the approval gate sits, the blast radius, and how you verify. Every level is grounded in real catalog entries.
Open →Which agent do you need?
Describe the job, where it should live, and how much leash you're willing to give. The quiz maps your answers onto the catalog's own vocabulary — category, interfaces, capability tags — and scores every live system.
Open →Same task, different anatomy
The fastest way to see what 'kind of agent' means: hold the task constant and watch the trace change. A broken test, a flight booking, a refund call — executed step by step as a CLI agent, a cloud agent, a computer-use agent, a voice agent.
Open →Anatomy of an MCP call
One tool call end to end: the model emits a structured request, it crosses the JSON-RPC boundary to a separate server, executes where the model can't reach, and the result re-enters context. The boundary is the point.
Open →