-
Improving the Efficiency of Language Agent Teams with Adaptive Task Graphs
Authors:
Elizabeth Mieczkowski,
Alexander Ku,
Tiwalayo Eisape,
Dilip Arumugam,
John Matters,
Katherine M. Collins,
Ilia Sucholutsky,
Thomas L. Griffiths
Abstract:
Large language models (LLMs) are increasingly deployed in teams, yet existing coordination approaches often occupy two extremes. Highly structured methods rely on fixed roles, pipelines, or task decompositions assigned a priori. In contrast, fully unstructured teams enable adaptability and exploration but suffer from inefficiencies such as error propagation, inter-agent conflicts, and wasted resou…
▽ More
Large language models (LLMs) are increasingly deployed in teams, yet existing coordination approaches often occupy two extremes. Highly structured methods rely on fixed roles, pipelines, or task decompositions assigned a priori. In contrast, fully unstructured teams enable adaptability and exploration but suffer from inefficiencies such as error propagation, inter-agent conflicts, and wasted resources (measured in time, tokens, or file operations). We introduce Language Agent Teams for Task Evolution (LATTE), a framework for coordinating LLM teams inspired by distributed systems, where processors must operate under partial observability and communication constraints. In LATTE, a team of agents collaboratively construct and maintain a shared, evolving coordination graph which encodes sub-task dependencies, individual agent assignment, and the current state of sub-task progress. This protocol maintains consistency while empowering agents to dynamically allocate work, adapt coordination, and discover new tasks. Across multiple collaborative tasks and a variety of base models, we demonstrate how LATTE reduces token usage, wall-clock time, communication, and coordination failures (e.g. file conflicts and redundant outputs) while matching or exceeding the accuracy of standard designs including MetaGPT, decentralized teams, top-down Leader-Worker hierarchies, and static decompositions.
△ Less
Submitted 7 May, 2026;
originally announced May 2026.
-
Language Model Teams as Distributed Systems
Authors:
Elizabeth Mieczkowski,
Katherine M. Collins,
Ilia Sucholutsky,
Natalia Vélez,
Thomas L. Griffiths
Abstract:
Large language models (LLMs) are growing increasingly capable, prompting recent interest in LLM teams. Yet, despite increased deployment of LLM teams at scale, we lack a principled framework for addressing key questions such as when a team is helpful, how many agents to use, how structure impacts performance -- and whether a team is better than a single agent. Rather than designing and testing the…
▽ More
Large language models (LLMs) are growing increasingly capable, prompting recent interest in LLM teams. Yet, despite increased deployment of LLM teams at scale, we lack a principled framework for addressing key questions such as when a team is helpful, how many agents to use, how structure impacts performance -- and whether a team is better than a single agent. Rather than designing and testing these possibilities through trial-and-error, we propose using distributed systems as a principled foundation for creating and evaluating LLM teams. We find that many of the fundamental advantages and challenges studied in distributed computing also arise in LLM teams, highlighting the rich practical insights that can come from the cross-talk of these two fields of study.
△ Less
Submitted 12 March, 2026;
originally announced March 2026.
-
Partner Modelling Emerges in Recurrent Agents (But Only When It Matters)
Authors:
Ruaridh Mon-Williams,
Max Taylor-Davies,
Elizabeth Mieczkowski,
Natalia Velez,
Neil R. Bramley,
Yanwei Wang,
Thomas L. Griffiths,
Christopher G. Lucas
Abstract:
Humans are remarkably adept at collaboration, able to infer the strengths and weaknesses of new partners in order to work successfully towards shared goals. To build AI systems with this capability, we must first understand its building blocks: does such flexibility require explicit, dedicated mechanisms for modelling others -- or can it emerge spontaneously from the pressures of open-ended cooper…
▽ More
Humans are remarkably adept at collaboration, able to infer the strengths and weaknesses of new partners in order to work successfully towards shared goals. To build AI systems with this capability, we must first understand its building blocks: does such flexibility require explicit, dedicated mechanisms for modelling others -- or can it emerge spontaneously from the pressures of open-ended cooperative interaction? To investigate this question, we train simple model-free RNN agents to collaborate with a population of diverse partners. Using the `Overcooked-AI' environment, we collect data from thousands of collaborative teams, and analyse agents' internal hidden states. Despite a lack of additional architectural features, inductive biases, or auxiliary objectives, the agents nevertheless develop structured internal representations of their partners' task abilities, enabling rapid adaptation and generalisation to novel collaborators. We investigated these internal models through probing techniques, and large-scale behavioural analysis. Notably, we find that structured partner modelling emerges when agents can influence partner behaviour by controlling task allocation. Our results show that partner modelling can arise spontaneously in model-free agents -- but only under environmental conditions that impose the right kind of social pressure.
△ Less
Submitted 27 October, 2025; v1 submitted 22 May, 2025;
originally announced May 2025.
-
Predicting Multi-Agent Specialization via Task Parallelizability
Authors:
Elizabeth Mieczkowski,
Ruaridh Mon-Williams,
Neil Bramley,
Christopher G. Lucas,
Natalia Velez,
Thomas L. Griffiths
Abstract:
When should we encourage specialization in multi-agent systems versus train generalists that perform the entire task independently? We propose that specialization largely depends on task parallelizability: the potential for multiple agents to execute task components concurrently. Drawing inspiration from Amdahl's Law in distributed systems, we present a closed-form bound that predicts when special…
▽ More
When should we encourage specialization in multi-agent systems versus train generalists that perform the entire task independently? We propose that specialization largely depends on task parallelizability: the potential for multiple agents to execute task components concurrently. Drawing inspiration from Amdahl's Law in distributed systems, we present a closed-form bound that predicts when specialization improves performance, depending only on task concurrency and team size. We validate our model on two standard MARL benchmarks that represent opposite regimes -- StarCraft Multi-Agent Challenge (SMAC, unlimited concurrency) and Multi-Particle Environment (MPE, unit-capacity bottlenecks) -- and observe close alignment between the bound at each extreme and an empirical measure of specialization. Three follow-up experiments in Overcooked-AI demonstrate that the model works in environments with more complex spatial and resource bottlenecks that allow for a range of strategies. Beyond prediction, the bound also serves as a diagnostic tool, highlighting biases in MARL training algorithms that cause sub-optimal convergence to specialist strategies with larger state spaces.
△ Less
Submitted 17 September, 2025; v1 submitted 19 March, 2025;
originally announced March 2025.
-
Arithmetic of triangles
Authors:
Edward Mieczkowski
Abstract:
In this paper, we consider a set of similar triangles with parallel sides, along with a set of points in the plane. It turns out that the set $\mathbb{R}_2= \{\pm <x >=\pm (x^2,x,1); x\in\mathbb{R} \}$ describes this set of triangles quite well. The set $\mathbb{R}_2$ is a subset of the ring $\mathbb{R}^3=\mathbb{R}\times\mathbb{R}\times\mathbb{R}= \{ (x,y,z) ; x,y,z\in\mathbb{R} \}$ with addition…
▽ More
In this paper, we consider a set of similar triangles with parallel sides, along with a set of points in the plane. It turns out that the set $\mathbb{R}_2= \{\pm <x >=\pm (x^2,x,1); x\in\mathbb{R} \}$ describes this set of triangles quite well. The set $\mathbb{R}_2$ is a subset of the ring $\mathbb{R}^3=\mathbb{R}\times\mathbb{R}\times\mathbb{R}= \{ (x,y,z) ; x,y,z\in\mathbb{R} \}$ with addition and multiplication defined coordinate-wise.
The set $\mathbb{R}_2$ is equipped with two operations. Multiplication is inherited from the ring $\mathbb{R}^3$, while addition is a ternary operation that represents homothety and translation of elements in $\mathbb{R}_2$.
However, the defined addition has its limitations. It turns out that, within this framework, the reduction of terms with different signs is not always possible. This leads to the distinction between an equation that is true in the arithmetic sense and one that is true in the geometric sense.
A novel form of addition in $\mathbb{R}_2$ leads to intriguing properties of multiplication in $\mathbb{R}_2$, which are examined in a dedicated chapter.
In the next section we use the construction of adding to describe the dissection of the triangle into 15 triangles of different sides.
In the final two sections, we consider a set of two kinds of vectors, along with a set of points on the line. The set $\mathbb{R}_1= \{\pm <x >=\pm (x,1); x\in\mathbb{R} \}$ describes this set vectors quite well and it is a one-dimensional reduction of the set $\mathbb{R}_2$.
△ Less
Submitted 16 September, 2025; v1 submitted 23 March, 2013;
originally announced March 2013.
-
The arithmetic of simplices
Authors:
Edward Mieczkowski
Abstract:
This paper continues the study initiated in "The aithmetic of Triangles." We begin by examining a set of similar tetrahedra with parallel sides, together with a set of points in three-dimensional space. It turns out that the set $\mathbb{R}_3= \{\pm <x >=\pm (x^3,x^2,x,1); x\in\mathbb{R} \}$ effectively characterizes this family of tetrahedra. The set $\mathbb{R}_3$ is a subset of the ring…
▽ More
This paper continues the study initiated in "The aithmetic of Triangles." We begin by examining a set of similar tetrahedra with parallel sides, together with a set of points in three-dimensional space. It turns out that the set $\mathbb{R}_3= \{\pm <x >=\pm (x^3,x^2,x,1); x\in\mathbb{R} \}$ effectively characterizes this family of tetrahedra. The set $\mathbb{R}_3$ is a subset of the ring $\mathbb{R}^4 = \mathbb{R} \times \mathbb{R} \times \mathbb{R} \times \mathbb{R} = \{ (x, y, z, w) ; x, y, z, w \in \mathbb{R} \}$, with addition and multiplication defined component-wise. The set $\mathbb{R}_3$ supports two operations. Multiplication is inherited directly from the ring $\mathbb{R}^4$, while addition is a four-argument operation that reflects geometric transformations such as homothety and translation of elements in $\mathbb{R}_3$. A novel form of addition in $\mathbb{R}_3$ leads to intriguing properties of multiplication in $\mathbb{R}_3$, which are examined in a dedicated chapter. We then generalize this approach to sets of $k$-dimensional similar simplices with parallel sides, along with corresponding sets of points in $k$-dimensional space.
△ Less
Submitted 16 September, 2025; v1 submitted 6 April, 2012;
originally announced April 2012.