A practical lab · about 19 minutes
Interactive learning lab
How AI works · guide 7 of 8
The bounded agent execution lab
0/5 core signals
Saved only on this device
Agent lab · BriefingScene 1 / 6
An agent is a model inside an action loop
The model chooses; tools execute; observations change the next choice.
An agent is a model inside an action loop ready.
Predict
Inspect
Verify
The minimum agent combines an LLM, tools and a loop until a stop condition. This creates capability and new failure modes: wrong tools, malformed arguments, loops, premature completion, permission abuse and accumulating cost.
Tool schemas define possible actions.
Observations become new context.
Stop conditions and permissions are product logic.