Hyperagents
Jenny Zhang, Bingchen Zhao, Wannan Yang, Jakob Foerster, Jeff Clune, Minqi Jiang, Sam Devlin, Tatiana Shavrina
2026-03-23
Summary
This paper introduces a new type of artificial intelligence system called a 'hyperagent' that can not only solve problems but also improve its *own* ability to solve problems and even improve *how* it improves. It builds on previous work but overcomes a key limitation of those systems.
What's the problem?
Current AI systems that try to improve themselves have a big weakness: the way they improve is usually pre-programmed by humans. This means there's a limit to how much better they can get, because the improvement process itself isn't evolving. Earlier systems also struggled to apply improvements made in one area, like coding, to other areas. Essentially, getting better at coding didn't automatically make them better at improving themselves in general.
What's the solution?
The researchers created 'hyperagents' which combine a problem-solving part ('task agent') with a self-improvement part ('meta agent') into a single, editable program. The crucial part is that the self-improvement process *itself* can be modified. This allows the AI to improve not just its ability to do tasks, but also its ability to *learn how to learn*. They built a specific version called DGM-Hyperagents, which doesn't rely on improvements in one area automatically translating to improvements in others, making it more flexible.
Why it matters?
This work is important because it represents a step towards AI systems that can truly learn and improve on their own, without constant human intervention. The hyperagents don't just find better solutions; they get better at *finding* better solutions. This could lead to AI that can tackle increasingly complex problems and adapt to new situations more effectively, potentially accelerating progress in many fields.
Abstract
Self-improving AI systems aim to reduce reliance on human engineering by learning to improve their own learning and problem-solving processes. Existing approaches to self-improvement rely on fixed, handcrafted meta-level mechanisms, fundamentally limiting how fast such systems can improve. The Darwin Gödel Machine (DGM) demonstrates open-ended self-improvement in coding by repeatedly generating and evaluating self-modified variants. Because both evaluation and self-modification are coding tasks, gains in coding ability can translate into gains in self-improvement ability. However, this alignment does not generally hold beyond coding domains. We introduce hyperagents, self-referential agents that integrate a task agent (which solves the target task) and a meta agent (which modifies itself and the task agent) into a single editable program. Crucially, the meta-level modification procedure is itself editable, enabling metacognitive self-modification, improving not only the task-solving behavior, but also the mechanism that generates future improvements. We instantiate this framework by extending DGM to create DGM-Hyperagents (DGM-H), eliminating the assumption of domain-specific alignment between task performance and self-modification skill to potentially support self-accelerating progress on any computable task. Across diverse domains, the DGM-H improves performance over time and outperforms baselines without self-improvement or open-ended exploration, as well as prior self-improving systems. Furthermore, the DGM-H improves the process by which it generates new agents (e.g., persistent memory, performance tracking), and these meta-level improvements transfer across domains and accumulate across runs. DGM-Hyperagents offer a glimpse of open-ended AI systems that do not merely search for better solutions, but continually improve their search for how to improve.