Agents

How a model asks for tools, loops on the result, and when a person still decides.

Five cards on this track are live. Concept cards are free, including the ones that are not written yet. Names without a card are not links. Practice for this track is the Agents pack, and that pack is coming soon.

Agent loop Live Free card
A cycle of model step, tool result, and another model step, until a stop.
Tool calling Live Free card
The model asks your code to run a named function, then reads the result.
Tool design/MCP Live Free card
How a tool is named, described, and exposed, including over MCP.
Context engineering Live Free card
Deciding what goes into the window for this step, and what stays out.
Memory Coming soon Free card
State kept across turns or sessions so the loop does not start from zero.
Planning Coming soon Free card
Breaking a job into steps before or during the loop.
Prompt injection Live Free card
Untrusted text that tries to override the instruction or the tool rules.
Permissions & human-in-the-loop Coming soon Free card
What the agent may do alone, and what waits for a person.
Guardrails Coming soon Free card
Checks that block or reshape an action or a reply before it lands.
Failure modes Coming soon Free card
How an agent loop stalls, loops, or acts on a bad tool result.
Agent evaluation Coming soon Free card
Scoring an agent on tasks, including whether it used tools well.
Multi-agent Coming soon Free card
More than one agent, with a handoff or a shared task.

Agents pack Pricing