Open and Efficient Agentic AI for Resource-Constrained Environments

About

Recent advances in foundation models have enabled increasingly capable agentic AI systems that perform reasoning, planning, tool use, workflow automation, and autonomous decision-making. However, much of this progress relies on frontier-scale proprietary models that require substantial computational resources, specialized infrastructure, and access to commercial APIs. These requirements create significant barriers for many researchers, startups, enterprises, and public-sector organizations operating under limited resources.

At the same time, rapid progress in open-weight foundation models, small language models, retrieval systems, memory architectures, test-time scaling, and multi-agent collaboration has created new opportunities for building capable AI systems without relying exclusively on frontier-scale models. A fundamental research question emerges: can algorithmic and system-level innovations compensate for model scale and enable effective agentic AI under limited computational budgets?

This OE-Agent workshop aims to bring together researchers and practitioners working on open and efficient agentic AI. The workshop will focus on methods that improve accessibility, affordability, privacy, deployability, and sustainability while maintaining strong capabilities. By highlighting emerging research directions and practical deployment challenges, the workshop seeks to foster collaboration across academia, industry, and government.

Motivation and Scope

Agentic AI has emerged as one of the most important directions in machine learning. Recent systems demonstrate impressive capabilities in reasoning, planning, memory utilization, tool interaction, and autonomous decision-making through the integration of large language models with memory, retrieval, planning, and external tools. These systems have the potential to transform scientific discovery, software engineering, healthcare, education, manufacturing, finance, and public services.

Despite these advances, the majority of state-of-the-art agentic systems are built upon frontier-scale proprietary foundation models. Training and deploying such systems often requires access to large computational budgets, specialized hardware, and commercial APIs. These requirements create substantial barriers for researchers and organizations operating under limited resources.

Many universities, startups, and public-sector organizations possess strong technical expertise but lack access to frontier-scale infrastructure. Consequently, there is growing interest in developing open, efficient, and deployable agentic systems that can operate effectively using open-weight models, limited computational budgets, and local infrastructure.

Recent progress in retrieval-augmented generation, memory systems, test-time scaling, tool use, model compression, distillation, and multi-agent collaboration suggests alternative pathways toward capable AI systems. These advances motivate a central question: can system-level innovations compensate for model scale and enable practical agentic AI in resource-constrained environments?

Topics of Interest

We invite submissions on topics including, but not limited to:

• Open-weight foundation models and small language models for agentic AI
• Efficient reasoning, planning, and test-time scaling for agents
• Memory, retrieval, tool use, and multi-agent collaboration
• Resource-constrained development and deployment, privacy-preserving AI, and edge agents
• Benchmarking and evaluation of open and efficient agentic systems
• Applications of open and efficient agents in various domains such as science, industry, and society

Submission Guidelines

We welcome both mature research contributions and promising early-stage work aligned with the workshop theme:

• Full papers — up to 8 pages, excluding references and appendices.
• Short papers — up to 4 pages, reporting preliminary results, emerging ideas, or position statements.

Submissions must be anonymized and formatted using the official ACML 2026 LaTeX template and style files (see the ACML 2026 Call for Papers for details). Each submission will receive at least two reviews, assessed on relevance, originality, technical quality, clarity, reproducibility, and potential impact. Particular attention will be given to work that advances open and efficient agentic AI under realistic computational, deployment, or resource constraints.

Submissions are handled through OpenReview: ACML 2026 OE-Agent Workshop submission portal.

Accepted papers will be presented as oral presentations or posters.

Important Dates

Program

Half-day workshop at ACML 2026. Exact times are to be updated once ACML announces the workshop timeslot.

Speakers and Panellists

Hajime Hotta

PhD, Director, Hajime Institute

Minh Tri Nguyen

PhD, Data Scientist, Commonwealth Bank

Patanamon Thongtanunam

Associate Professor, University of Melbourne

Nir Lipovetzky

Associate Professor, University of Melbourne

Invited speakers also join the panel discussion: Can Open and Efficient Agents Rival Frontier Models?

Organizers

Hung Le

A2I2, Deakin University

Haripriya Harikumar

Centre for AI Fundamentals, University of Manchester

Khanh-Binh (Calvin) Nguyen

A2I2, Deakin University

Dung Nguyen

A2I2, Deakin University

Manh Nguyen

A2I2, Deakin University

Long Hoang Dang

Posts and Telecommunications Institute of Technology

Contact

Sponsors

Sponsored by A2I2, Deakin University

Hosted by ACML 2026