Guibas, Agrawala co-author agentic-graphics taxonomy report
TL;DR
- A 27-page arxiv report with 14 authors, including Leonidas Guibas and Maneesh Agrawala, proposes a unified taxonomy for LLM-agent systems in computer graphics.
- The authors also build an 'agentic literature system' to help human researchers analyze existing and newly published graphics papers at scale.
- Listed open challenges include coordinating agents and tools, human-agent interfaces, long-horizon planning, and evaluating interactive graphics systems.
Fourteen researchers, including Leonidas Guibas and Maneesh Agrawala, posted a 27-page report on November 1, 2026 trying to put a common vocabulary around what they call agentic computer graphics.
"Computer graphics is built on code," the paper opens. "Large language model (LLM) agents can now generate graphics code and iteratively refine it through various feedback."
Two contributions, per the abstract. First, "a unified taxonomy for characterizing agentic systems across diverse computer graphics tasks." Second, "an agentic literature system that can assist human researchers in analyzing existing and newly published papers at scale." Agents being used to study agents.
The authors list their open challenges plainly: coordinating agents and tools across graphics tasks, designing interfaces and representations for human-agent collaboration, sustaining long-horizon planning and refinement, and reliably evaluating interactive graphics systems. The abstract does not name which specific graphics tasks the taxonomy covers, which LLMs power the literature system, or any accuracy numbers for it.
Originally reported by paper
Read the original article →Original headline: SIGGRAPH 2026 Survey Defines 'Agentic Computer Graphics' as a New Research Paradigm