The study evaluates four large language model agent architectures—single function‑calling, skill‑augmented, centralized multi‑agent orchestration, and triage swarm—across 13 enterprise standard operating procedures (SOPs). Results show no single architecture dominates; success varies with SOP complexity, and added structural complexity often incurs higher token usage and latency without significant performance gains. The research highlights that implementation quality, rather than the choice between single or multi‑agent designs, is the key determinant of task success.
arXiv:2607. 06906v1 Announce Type: new Abstract: Agentic AI development today runs on token maxing: buying capability with tokens -- longer reasoning traces, more turns, wider tool payloads, bigger replayed contexts -- so tokens per task grow faster than task value.
Agentic AI development today runs on token maxing: buying capability with tokens -- longer reasoning traces, more turns, wider tool payloads, bigger replayed contexts -- so tokens per task grow faster than task value. Falling per-token prices mask the pattern; total spend rises anyway.
arXiv:2610.35434v1 Announce Type: new Abstract: Recursive self-improvement (RSI), in which a system also improves the way it improves, is a long-standing aspiration on the path to AGI. Language agent...
arXiv:2610.14918v1 Announce Type: new Abstract: Evaluating an agent's ability to strictly satisfy diverse and procedurally complex requirements is essential to determining its practical reliability....
arXiv:2609. 37968v1 Announce Type: new Abstract: Advances in the coding capabilities of LLM agents allow them to inspect and modify their own instructions, tools, and execution procedures.
The paper investigates how users should set the reasoning‑effort level in large language models, a parameter that trades off inference cost for potential performance gains. Experiments on proprietary and open‑weight models across four benchmarks show that higher effort does not consistently improve accuracy, can be up to 42× more expensive on easy tasks, and alters agent behavior by increasing verification and search steps. The authors propose a simple escalation strategy—starting at the lowest effort and only increasing after a verified failure—that matches or outperforms fixed high‑effort settings while reducing cost.
arXiv:2608. 08466v1 Announce Type: new Abstract: Modern LLM agents are often improved by modifying prompts, tools, or workflows manually, while the executable scaffold surrounding the model---the \emph{harness}---is typically treated as a fixed artifact after deployment.
arXiv:2607. 05297v1 Announce Type: new Abstract: Recent LLM agents tackle increasingly long-horizon, open-ended tasks, and external skills, reusable procedural knowledge supplied to the agent, further extend this capability.
ReTrace converts a standard operating procedure (SOP) and a handful of logged agent runs into a reusable, planner‑free workflow that uses code, not an LLM, to select the next step. By extracting the conditions behind each tool call and merging them into a compact rule document, ReTrace produces an executable workflow that can be compiled for various orchestration platforms. In tests, ReTrace achieves high completion rates (up to 93.8 %) with significantly lower cost and faster execution (1.3 s vs. 170 s) compared to the original LLM agent.
The paper investigates how scaling a team of small language‑model agents affects performance across different orchestration architectures. By testing eight architectures on five short‑answer benchmarks and an executable‑code benchmark, it finds that team scaling yields large gains on arithmetic word‑problem tasks but only modest improvements on multiple‑choice and code generation tasks, with no single architecture dominating all tasks. The authors explain these patterns using a generate‑transform decomposition that separates coverage and transformation effects, showing that arithmetic tasks benefit from both coverage and critic‑guided transformation, while other tasks are limited by saturation or poor conversion.
The paper introduces BiPA, a method to evaluate skill optimizers that rewrite natural‑language procedures for agents performing long‑horizon tasks. BiPA replays the optimizer’s acceptance rule on identical candidates to isolate selection effects and records each episode as a bipartite graph of turns and read text, revealing that learned skills only marginally outperform hand‑written seeds and that many gains vanish on test data. The study finds optimizers add extra input tokens and turns, and proposes four design principles for more reliable skill editing.
The paper introduces BiPA, a method for evaluating skill optimizers that rewrite natural‑language procedures for agents. By replaying the optimizer’s acceptance rule on identical candidates and recording episodes as bipartite graphs of turns and read text, BiPA isolates the effect of selection from the cost of added input tokens. Experiments on three optimizers, three benchmarks, and two models show that learned skills perform only marginally better than hand‑written seeds, and that apparent dev‑set gains often disappear on test, revealing that optimizers tend to add redundant rules that increase token usage.