LLM Agents: A Survey of Architectures & Designs (2022–2026)
LLM Agents: A Survey (2022–2026)
A thorough review of the landscape of LLM-based agent architectures, designs, and the evolving ecosystem — covering papers, preprints, frameworks, blog posts, and open-source tools.
Compiled March 2026 · View on GitHub
What Is This?
This survey maps the landscape of Large Language Model (LLM) based agents — systems where LLMs serve not just as text generators, but as reasoning engines that can plan, use tools, access memory, and take actions in the world.
From 2022 to 2026, the field has exploded. This guide aims to be a navigable map: organized by theme, covering both the academic literature and the practitioner ecosystem.
How to Use This Survey
- New to agents? Start with Taxonomy, then Foundations
- Researcher? Each section has structured paper entries with links and summaries
- Practitioner? Check Multi-Agent Systems for frameworks and 2024–2026 for the current ecosystem
- Looking for resources? See Resources & Links for a curated reading list
Scope & Methodology
This survey covers:
- Papers & preprints (arXiv, ACL, NeurIPS, ICML, ICLR, etc.)
- Technical blog posts from researchers and practitioners
- Open-source frameworks and tools (GitHub)
- Time range: 2022 through March 2026
The focus is on LLM-based agents — systems using large language models as a core reasoning component. We touch on embodied/robotics agents when they are directly relevant, but the primary focus is on digital, language-model-centric work.
This survey was compiled in March 2026. The field moves fast — if you spot a missing reference, open an issue.