Agentic in-context learning – Definition & Meaning

Definition of agentic in-context learning

: a form of in-context learning in which an AI agent actively chooses actions, retrieves or generates useful demonstrations, observes the results, and updates the context it uses to solve a task without changing the model's weights
- The research agent used agentic in-context learning to search for a missing example before attempting the final proof.
- With agentic ICL, the model decided which demonstrations to retrieve instead of receiving a fixed few-shot prompt.

Origin & history

first known use2025

The phrase appeared in 2025 research on journalism-guided in-context learning and broadened in 2026 work that treated an agent's interaction loop as a learning process. Papers published in October 2026 used the term for agents that actively acquire demonstrations or determine what evidence enters their context next.

Editorial references

Sources & further reading

3 sources

Journalism-Guided Agentic In-context Learning for News Stance Detection Peer Reviewed Research Paper · Association for Computational Linguistics · 2025
ActICL: Post-Training for Agentic Multimodal In-Context Learning Research Preprint · arXiv · 2026-10-14
Agentic In-Context Learning: What Algorithms Can Transformers Learn in Closed-Loop Interaction? Research Preprint · arXiv · 2026-10-16

Test yourself

Which of these is the meaning of agentic in-context learning?

Cite this entry

"Agentic in-context learning." AI Dictionary, Dadgogo, https://dadgogo.com/dictionary/agentic-in-context-learning/. Accessed 19 Oct. 2026.