Fri 13 Nov
LiveGet it free
ZOTPAPER

News without the noise

← All concepts

agent-driven simulation

Think of a simulation as a model of some world, like a game, a market, or a robot's surroundings. Instead of a human coding every rule and behavior, an agent-driven simulation lets an AI agent build, change, or run parts of that model itself. These agents can write code, act as simulated people or businesses, explore different scenarios, and learn from the results. You tell the agent what you want to simulate, and it figures out the details dynamically, making the process faster and more flexible than hand-crafting everything.

Across these papers, the common thread is that agents are taking over tasks humans used to do in simulation: generating realistic user behaviors, extending server simulators, teaching robots physical skills, creating benchmarks, and even managing entire companies. They often match or beat traditional methods in accuracy and speed, while also handling complexity that manual simulation struggles with. One surprising finding is that agents can build excellent simulations but still make poor decisions inside them, suggesting a practical split between offline construction and real-time action.

Papers
39
Released code
3
First seen
July 2026
Latest
Nov 2026

31 papers in the last two months, against 6 in the two before.

Start here

Three papers to read first, in the order they came out
  1. 1AI agents build a Hearthstone Battlegrounds bot that rivals top players

    A self-improving team of LLM research agents built a Hearthstone Battlegrounds AI that beats top players, while the LLMs themselves play far worse.

    Oct 2026
  2. 2Language models improve psychiatric interviews after training with simulated patients

    Training a language model with multi-turn reinforcement learning in simulated psychiatric intakes improves its diagnostic, communication, and safety performance over its base model.

    Oct 2026
  3. 3A small AI agent trained on cloned websites matches bigger web agents

    A suite of 153 self-hosted replica websites with 3,518 tasks trains small web agents that then handle unseen tasks and match leading open agents.

    Nov 2026

How fast it is moving

JunJulAugSeptOctNov so far

Often studied with

agentic evaluationMeasuring AI agents' performance in complex tasks, highlighting failures in planning, monitoring, and real9 togethercoding agentAn AI-powered software system that autonomously writes, modifies, or extends source code based on high-level instructions.9 togetherlarge language modelA neural network trained on massive text corpora to generate coherent text and perform diverse language tasks.8 together

Who is working on it

Peking University 3Shanghai Jiao Tong University 3University of Illinois Urbana-Champaign 2Microsoft 2Harvard University 2Alibaba Group 2

The papers

Most central to this idea first, not most recent.

Concepts are extracted from each paper and reused across the corpus, so this page grows on its own as the desk reads.