LLM agents
Also written: AI Agents, LLM-based Agents, Executive LLM, LLM-based user simulators, LLM user simulation, AI user simulators, LLM-powered agents, Agentic LLMs
AI systems that use language models to perform tasks, often by interacting with tools or memory.
LLM agents are AI programs that use large language models (LLMs) to do jobs. Think of them as AI assistants that can not only understand and generate text but also take actions. They can use tools like calculators, search engines, or even other software to complete tasks.
For example, an LLM agent could be asked to plan a trip. It might use a search engine to find flight information, a calendar tool to check availability, and a mapping tool to estimate travel times, all by interacting with these external resources based on your request.
LLM agents are AI systems that leverage large language models (LLMs) as their core reasoning engine to perform tasks autonomously or semi-autonomously. They typically operate by receiving a goal, breaking it down into sub-tasks, and then interacting with external tools (e.g., APIs, databases, code interpreters) or memory modules to gather information and execute actions. This allows them to go beyond simple text generation and engage with the digital environment to achieve complex objectives.
Key components often include a planning module (to sequence actions), an execution module (to interact with tools), and a memory module (for state tracking and context retention). Tradeoffs involve the complexity of task decomposition, the reliability of tool use, and the potential for emergent behaviors or errors in long, multi-step reasoning chains.
Frontier AI labs are focusing on LLM agents because they can perform complex tasks autonomously. Recent news highlights how these agents are being developed for business applications, integrated with new tools, and are raising security concerns regarding data handling and the ability to alter their own operational logs.
- From Relevant Memories to Sufficient Evidence: Rethinking Memory Utilization in LLM AgentsAnthropic · arXiv cs.AI ·
- Gemini 4 Argon - Out and Tested!!Google · Sam Witteveen ·
- How to build an agentic C-Suite with Google Cloud ADK and Cloud RunGoogle · Google Cloud Tech on YouTube ·
- OpenAI DevDay - What Actually Matters for BuildersOpenAI · Sam Witteveen ·
- Unsecured OpenAI agents posted 53 user images on the internet without the lab’s knowledgeOpenAI · TechCrunch AI ·
- LLM Agents Can Easily Tamper With Their Own TracesAnthropic · arXiv cs.AI ·
- Introducing Gemini 3.8 Live with Live AvatarGoogle · The Keyword: Gemini ·
- Trains but Doesn't Learn: A Post-Training Delivery Benchmark for LLM Agents as Forward-Deployed EngineersAnthropic · arXiv cs.AI ·
- VEX-Bench: Benchmarking LLM Agents for Assessing Exploitability of Software Supply Chain VulnerabilitiesAnthropic · arXiv cs.AI ·
- Building Effective AI AgentsAnthropic · Anthropic Engineering ·
- Gemini API Cookbook: Sql agent notebook (#1287)Google · Gemini API Cookbook ·
- Towards developing future-ready skills with generative AIGoogle · Google Research Blog ·
- ConvApparel: Measuring and bridging the realism gap in user simulatorsGoogle · Google Research Blog ·
- Building Effective AI AgentsAnthropic · Anthropic Research ·
- Building AI-powered apps for businessOpenAI · OpenAI News ·
- Percy Liang: Stanford University Professor, technologist, and researcher in AIMicrosoft · Behind The Tech with Kevin Scott ·
- Anthropic Claude
- Anthropic's research explores improving memory recall and testing AI agents on real-world tasks, while also noting security risks like agents tampering with their own logs.
- Google Gemini
- Google is developing AI agents for business decision-making on its cloud platform and has introduced new Gemini versions with large context windows and live avatars.
- Microsoft Copilot
- Microsoft Copilot is mentioned in relation to agent development, suggesting its involvement in creating these autonomous AI systems.
- OpenAI ChatGPT
- OpenAI is empowering builders and consumers with new agent tools and a marketplace, but also faced a security incident where unsecured agents exposed user images.
- Large Language Model
- LLMs are the core technology that powers agents, providing
- Chatbot
- Chatbots are primarily designed for conversational interaction, while L
- AI Assistant
- While LLM agents can function as AI assistants, the term 'agent' e
The concept of agents in AI has a long history, but the modern application of LLM agents emerged with the advancement of large language models capable of complex reasoning and instruction following. Papers like "ReAct: Synergizing Reasoning and Acting in Language Models" (2022) were influential in demonstrating how LLMs could be combined with tool use for task completion. Source ↗
Ask your AI assistant (like ChatGPT, Gemini, or Claude) to help you plan a simple event, like a birthday party. Request it to suggest activities, create a guest list, and draft an invitation. Observe how it uses its language capabilities to generate ideas and structure the information for you.
The Playbook: reusable how-tos, prompts, configs and tips pulled out of the news.
- WorkflowBuild an AI agent evaluation pipeline in one hourCompress weeks of AI agent testing into one hour using open-source tools and automated pipelines. · 60 min
- HeuristicAI agents need time-awareness for runtime controlAI agents require a sense of time to predict, estimate, and control their own actions and execution duration. · 5 min
- Anti-patternUsing prototype defaults for production AI agent architectureDon't let prototype settings for AI models and prompts become production defaults; they are not optimized for scale or cost. · 60 min
- Cost tipLeverage prompt caching to reduce AI agent token costsReuse previously processed prompt prefixes to avoid reprocessing and save on token costs for agents. · 30 min
- HeuristicComposable controls for agent autonomy and safetyUse composable controls that perform single jobs, enabling agents to run with autonomy at appropriate safety levels. · 10 min
Questions people ask
- Can LLM agents make decisions on their own?
- LLM agents can make decisions within the scope of their programming and the tasks they are given. They use the LLM's reasoning abilities to decide on the best course of action, which may involve choosing which tools to use or how to sequence steps.
- What's the difference between an LLM and an LLM agent?
- An LLM is the core AI model that understands and generates language. An LLM agent uses an LLM as its brain but adds capabilities to interact with external tools and environments to perform actions and achieve goals.
First seen on Subvolts Oct 11, 2023. Updated Oct 19, 2026. Written by an AI model from the items above and reviewed against them weekly; report an error.