NEW RUNTIME

Topic hub

Tool USE

A New Runtime topic hub collecting signals, patterns, field notes, and public sources about tool use.

Short answer

  • Tool USE is tracked here as an evidence-linked topic, not as a static glossary entry.
  • The page connects raw observations to pattern hypotheses, longer analysis, and public sources.
  • Use it as the canonical landing page before drilling into individual records.

Field notes

Longer analysis

1 notes

  1. Agent APIs Need Fewer Magic Tricks and More Facts

    Agent-first APIs should return explicit fields, precise errors, raw facts, and traceable metadata so models can repair tool calls deterministically instead of guessing world state.

Raw signals

Recent observations

6 signals

  1. Pydantic AI loads capabilities only when needed

    A capability can bundle instructions, tools, model settings, and hooks while exposing only a compact description until activation.

  2. DoorDash Assistant Builds the Cart Instead of Giving Advice

    The assistant combines retrieval, personalization, and transaction tools so the product result is an editable action state, not prose.

  3. Agent Evals Become a Discipline Separate from Model Evals

    Reliable agents require evaluation of trajectories, tools, state, recovery, and outcomes rather than grading only the final answer.

  4. Google Search Adds Agentic Action to Discovery

    Search is extending from retrieval and answers into tasks, tool use, and guided execution inside a familiar high-traffic interface.

  5. A Personal Agent OS Emerges from Small CLI Tools

    A suite of narrow command-line adapters gives an agent composable access to messaging, media, GitHub, search, and local systems.

  6. Voice Agents Move from Conversation to Tool-Using Work

    OpenAI's voice stack combines low-latency speech with tool calls and state, making voice a working interface rather than a novelty chat.