Task

Agents

AI agents are autonomous software systems that use artificial intelligence to achieve goals and complete tasks on behalf of users, acting independently to perceive their environment, make decisions, and take actions without constant human intervention. They use advanced capabilities like reasoning, memory, planning, and learning, often leveraging large language models (LLMs) and other AI tools to interpret information and perform complex workflows across various industries. Current directions include: Agents are moving from short, prompt-scripted reasoning loops toward terminal-based systems that operate in real software environments. [ReAct](/paper/2210.03629) established the interleaved reasoning-and-action pattern; [Terminal-Bench 2.1](/benchmark/terminal-bench-2-1?task=agents) now emphasizes difficult multi-step command-line work with executable verification, where planning, tool use, recovery, and robust harnesses matter as much as model knowledge.; Long-horizon reasoning is becoming the central capability bottleneck. Widely reported benchmarks such as [BrowseComp](/benchmark/browsecomp?task=agents) and [Toolathlon](/benchmark/toolathlon?task=agents) reward persistent search and roughly 20-turn, multi-application workflows; their leading systems combine high reasoning effort with context compaction, code execution, programmatic tools, and multi-agent delegation to stay on task.; Personal agents are becoming persistent, user-specific systems rather than disposable chat sessions. The [OpenClaw project](https://openclaw.ai/) popularized always-on assistants connected to local tools and messaging surfaces, while the [Hermes Agent project](https://hermes-agent.nousresearch.com/) combines one cross-surface memory with learned skills, scheduling, and isolated subagents; [Claw-Eval](/benchmark/claw-eval?task=agents) reflects the resulting focus on completion, safety, and robustness across real-world interactions.