# Agents, Autonomy, and Tool Use

Recent AI moves from answering to acting: planning, using tools, and operating environments. This sub-article expands [Section 3 of Recent Topics](../recent.md) and complements [Agents & Tooling](agents-tooling.md).

## Agentic Workflows
- Models plan, call APIs, browse, write code, and act on users' behalf.
- **Computer-use agents** operate a desktop/browser like a human (Anthropic's computer use, OpenAI's Operator).

## Reasoning Models
- Chain-of-thought and "thinking" models (o1/o3-style) spend tokens on deliberation.

## Robotics
- **Vision-language-action models** (RT-2, Figure) connect LLMs to physical embodiment.

## Safety of Agents
- Autonomy raises new risks: unintended actions, persistent goals, and accountability gaps.
- See [AI Safety](ai-safety.md).

## Why It Matters
Agents turn language models into general-purpose tools that change how work and decisions happen.

## Sources
- Anthropic. *Computer use announcement* (2024).
- OpenAI. *Operator* (2025).

*Back to [Recent Topics](../recent.md).*