Agents across the ecosystem.

Read tool calling first to separate model proposals from application authority. The memory guide adds the question of which facts remain relevant across a conversation. The robotics guide then shows why physical execution needs its own controller, current-state checks, and independent stop behavior.

Use this collection to compare how the same design question changes across tasks. Read the source-backed explanation before adapting an example. Identify which assumptions apply to your environment, which information is unavailable, and which checks must remain in ordinary application code. A reference pattern is a starting point for that review, not evidence that a connected service or deployment already exists.

Keep a small set of difficult cases as you explore: an incomplete input, conflicting evidence, a changed permission, and an operation whose result is uncertain. Record what should happen in each case. When you finish a guide, revisit those expectations instead of judging the design only by its most fluent answer. The linked articles offer complementary perspectives, so a useful reading order follows the consequences of your task rather than the publication date alone.