Nandi, Subhrangshu, Arora, Hitesh, Bai, Yunfei, Patel, Udita, Shah, Chirag, Kwon, Hannah, Zhang, Wei, Chen, Weian, Li, Chloe Ruoqi, Sconfienza, Alberto, Goyal, Ayush, Lee, Yasmine, Xiong, Ming, Ramachandra, Prathap · Nov 7, 2026
arXiv cs.AI: Agent Architecture Cost-Success Trade-offs across Enterprise-Style SOPs
Markdown · JSON · Original source
The paper’s authors compare four agent architectures across 13 enterprise standard operating procedures (SOPs), holding the model, tasks and tools fixed while varying architecture. They report no universal best: success varies widely by SOP, and no architecture is statistically separable in aggregate. Added structure only shows a directional, not yet significant, trend toward helping on complex, branching SOPs. Coordinated designs use more tokens and take longer than a single agent, while their success gains are usually small and rarely significant; the authors therefore call the single agent the efficient default. But their results also make implementation quality central: high-quality skills improve success, degraded skills lower it, and a targeted redesign of a multi-agent system is the only arm reported to significantly beat the baseline. The authors’ argument is not simply that simpler agents win: extra architecture has to earn its cost on the particular workflow, and design quality can matter more than the single- versus multi-agent choice. The abstract does not establish how reliably these results transfer beyond the tested SOPs.
Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims.
Original source
cs.AI · Article
Source details
Original source date: Nov 7, 2026 (2026-11-07).
Date basis: Date basis not provided.
Added to Escape Meridian: 2026-11-12T03:31:32.274Z. First observed: 2026-11-11T05:06:28.241Z. Retrieved: 2026-11-12T03:30:47.222Z.
Referenced by
Recorded citations and public file attachments. A reference alone does not establish agreement or influence.
No public references recorded here yet.
Limits
The available evidence is the abstract; it does not provide the full methods, SOP descriptions, or detailed results needed to assess how the findings extend beyond the tested workflows.
Open question
What further evidence would show which complex, branching SOPs benefit enough from added architecture to justify its extra tokens and ?
Source-stated
The authors compare four agent architectures on 13 enterprise SOPs, holding the model, tasks and tools fixed while varying architecture.
Holding the model, tasks, and tools fixed and varying only the architecture, we compare four designs
Source-stated
The authors report that no architecture is statistically separable in aggregate, and that added structure’s apparent benefit on complex, branching SOPs is directional but not yet significant.
no architecture is statistically separable in aggregate, and added structure only plausibly pays off on complex, branching SOPs (a directional, not yet significant, trend).
Source-stated
The authors report that coordinated designs use more tokens and have higher latency than a single agent, with gains that are usually small and rarely significant; they call the single agent the efficient default.
Every coordinated arm (multi-agent or skill-augmented) spends $1.4$--$4.7\times$ the tokens and $2.6$--$5.8\times$ the latency of the single agent for gains that are usually small and rarely significant, making the single agent the efficient default
Source-stated
The authors report that skill quality and a targeted multi-agent redesign affect results; the redesigned arm is the only one reported to significantly beat the baseline.
high-quality skills do lift success (to $73.1\%$ from $67.4\%$) while degraded skills fall below it. (3)~\textbf{Implementation quality outweighs the single- versus multi-agent choice.} A plausible multi-agent implementation loses to the single agent, yet a targeted redesign of it becomes the only a
Source and contributor context
Nandi, Subhrangshu, Arora, Hitesh, Bai, Yunfei, Patel, Udita, Shah, Chirag, Kwon, Hannah, Zhang, Wei, Chen, Weian, Li, Chloe Ruoqi, Sconfienza, Alberto, Goyal, Ayush, Lee, Yasmine, Xiong, Ming, Ramachandra, Prathap · Article author
These are credits stated by the source; they do not establish a profile match or authorship of every passage.
Source media
No suitable media was found by the supported page extraction.
Terms and references
Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.
Latency
The delay between sending a request and receiving the result.
arXiv
arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Editorial provenance
Version 1. Publication ID: f5ab9893-1cb5-4f1f-82df-f16d0c64bd0e.
Prepared by gpt-6-luna; grounding review by gpt-6-luna.
Source check: passed by anthropic/claude-opus-5.5 on 2026-11-12T03:31:32.274Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: material_development.
Observation: 44df90786d280647000da0c8a28b651da3d497e51e19c65bbaef2f1d06c2499a. Source text hash: 7ce81b10df539bcc048cf52a2ca9a0f9ae2aea24cfb4afde98bfd8a4973d929c.
Revision and source-check history
Version 1 · 2026-11-12T03:31:32.274Z
arXiv cs.AI: Agent Architecture Cost-Success Trade-offs across Enterprise-Style SOPs
The paper’s authors compare four agent architectures across 13 enterprise standard operating procedures (SOPs), holding the model, tasks and tools fixed while varying architecture. They report no universal best: success varies widely by SOP, and no architecture is statistically separable in aggregate. Added structure only shows a directional, not yet significant, trend toward helping on complex, branching SOPs. Coordinated designs use more tokens and take longer than a single agent, while their success gains are usually small and rarely significant; the authors therefore call the single agent the efficient default. But their results also make implementation quality central: high-quality skills improve success, degraded skills lower it, and a targeted redesign of a multi-agent system is the only arm reported to significantly beat the baseline. The authors’ argument is not simply that simpler agents win: extra architecture has to earn its cost on the particular workflow, and design quality can matter more than the single- versus multi-agent choice. The abstract does not establish how reliably these results transfer beyond the tested SOPs.
Original source
cs.AI · Article
Source details
Original source date: Nov 7, 2026 (2026-11-07).
Date basis: Date basis not provided.
Added to Escape Meridian: 2026-11-12T03:31:32.274Z. First observed: 2026-11-11T05:06:28.241Z. Retrieved: 2026-11-12T03:30:47.222Z.
Limits
The available evidence is the abstract; it does not provide the full methods, SOP descriptions, or detailed results needed to assess how the findings extend beyond the tested workflows.
Open question
What further evidence would show which complex, branching SOPs benefit enough from added architecture to justify its extra tokens and ?
Source-stated
The authors compare four agent architectures on 13 enterprise SOPs, holding the model, tasks and tools fixed while varying architecture.
Holding the model, tasks, and tools fixed and varying only the architecture, we compare four designs
Source-stated
The authors report that no architecture is statistically separable in aggregate, and that added structure’s apparent benefit on complex, branching SOPs is directional but not yet significant.
no architecture is statistically separable in aggregate, and added structure only plausibly pays off on complex, branching SOPs (a directional, not yet significant, trend).
Source-stated
The authors report that coordinated designs use more tokens and have higher latency than a single agent, with gains that are usually small and rarely significant; they call the single agent the efficient default.
Every coordinated arm (multi-agent or skill-augmented) spends $1.4$--$4.7\times$ the tokens and $2.6$--$5.8\times$ the latency of the single agent for gains that are usually small and rarely significant, making the single agent the efficient default
Source-stated
The authors report that skill quality and a targeted multi-agent redesign affect results; the redesigned arm is the only one reported to significantly beat the baseline.
high-quality skills do lift success (to $73.1\%$ from $67.4\%$) while degraded skills fall below it. (3)~\textbf{Implementation quality outweighs the single- versus multi-agent choice.} A plausible multi-agent implementation loses to the single agent, yet a targeted redesign of it becomes the only a
Source and contributor context
Nandi, Subhrangshu, Arora, Hitesh, Bai, Yunfei, Patel, Udita, Shah, Chirag, Kwon, Hannah, Zhang, Wei, Chen, Weian, Li, Chloe Ruoqi, Sconfienza, Alberto, Goyal, Ayush, Lee, Yasmine, Xiong, Ming, Ramachandra, Prathap · Article author
These are credits stated by the source; they do not establish a profile match or authorship of every passage.
Source media
No suitable media was found by the supported page extraction.
Terms and references
Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.
Latency
The delay between sending a request and receiving the result.
arXiv
arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Editorial provenance
Version 1. Publication ID: f5ab9893-1cb5-4f1f-82df-f16d0c64bd0e.
Prepared by gpt-6-luna; grounding review by gpt-6-luna.
Source check: passed by anthropic/claude-opus-5.5 on 2026-11-12T03:31:32.274Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: material_development.
Observation: 44df90786d280647000da0c8a28b651da3d497e51e19c65bbaef2f1d06c2499a. Source text hash: 7ce81b10df539bcc048cf52a2ca9a0f9ae2aea24cfb4afde98bfd8a4973d929c.
passed · 2026-11-12T03:31:32.274Z · anthropic/claude-opus-5.5
Publication: f5ab9893-1cb5-4f1f-82df-f16d0c64bd0e
attribution — pass: The reading attributes the comparison, findings and the 'efficient default' conclusion to the paper's authors. That matches the arXiv abstract, which is written in the authors' voice. The reading presents no claims as established fact or as Escape Meridian's opinion.
Holding the model, tasks, and tools fixed and varying only the architecture, we compare four designs
summary — pass: The summary keeps all three takeaways, including the not-yet-significant trend and the point that implementation quality outweighs the single- versus multi-agent choice. It also keeps the authors' nuance that extra architecture must earn its cost. SOPs are explained in plain language. Terms like tokens, latency, skills and arm can be handled by tooltips and do not block understanding. The title is accurate, and the question follows from the directional finding without leading.
Implementation quality outweighs the single- versus multi-agent choice.
date — pass: The displayed original publication date of 2026-11-07 matches the v1 submission date on arXiv. The feed date of 2026-11-11 was correctly not used.
[Submitted on 7 Nov 2026]
support — pass: Each claim is grounded in exact abstract excerpts, and the hedges stay intact: 'directional, not yet significant' and 'usually small and rarely significant'. The limitation about how far the results transfer is correctly framed as unknown. There are no invented benefits, costs or forecasts.
Every coordinated arm (multi-agent or skill-augmented) spends $1.4$--$4.7\times$ the tokens and $2.6$--$5.8\times$ the latency of the single agent
media — pass: The selected media is the verified direct arXiv abstract page for this paper. It matches the source URL and is the most direct way to reach the original.
Challenge or correct this reading
Send a correction or an opposing source with this reading’s URL. Disagreement and uncertainty can remain unresolved.
arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Explore arXivThe delay between sending a request and receiving the result.
Definition referencearXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Explore arXivThe delay between sending a request and receiving the result.
Definition reference