When the Mythical Man-Month Meets Agentic Web Systems: Structured Coordination for Scalable Development
Authors: Zhiyao Wu, Zi Liang, Haibo Hu
Organizations: Hong Kong Polytechnic University, Hong Kong, China · Hong Kong Polytechnic University, PolyU Research Centre for Privacy and Security Technologies in Future Smart Systems, Hong Kong, China
As agentic systems are increasingly deployed to support complex Web tasks, multi-agent architectures offer greater parallelism, specialization, and task coverage. However, increasing the number of agents does not necessarily improve system performance, as communication, synchronization, redundant work, and coordination overhead may offset the benefits of additional agents. This presents a fundamental scaling challenge for agentic Web systems: how can agent populations be organized to balance task performance and coordination costs? To address this challenge, we systematically investigate the relationship between agent population size, per-agent token consumption, and coordination overhead in collaborative agentic systems. Experiments on complex Web-oriented software development tasks reveal a non-linear scaling pattern: per-agent token consumption initially decreases as the agent population grows, but eventually exhibits diminishing returns. Motivated by these findings, we introduce the Surgical Team Model (STM), a structured coordination framework that organizes agents through role specialization, constrained inter-agent communication, and coordinator-mediated task aggregation. Experiments on full-stack Web application development tasks show that STM achieves competitive task completion quality and comparable token consumption relative to representative baselines, while incurring higher execution latency.
Figures & tables
Figure 1. Per-agent token consumption (top of each panel) and total token consumption (bottom of each panel) under different agent populations. Panels: (a) e-commerce, (b) movie-ticket booking, (c) commodity inventory, (d) personal accounting, (e) library management, (f) equipment borrowing, (g) birthday reminder, (h) volunteer registration. Shaded bands on the per-agent curve mark the parallel-gain, coordination-efficient, and coordination-dominated regimes, and the marker indicates the empirical optimum n∗ . One legend is shared across panels. Eight task panels. Each panel has a U-shaped per-agent token curve on top and a rising total token curve on the bottom as the agent population increases, with a marker at the empirical optimum.
Figure 2. STM framework with structured interaction topology. STM framework with a structured interaction topology. The system adopts a hub-and-spoke architecture in which a central coordinator manages interactions with specialized agents, replacing unrestricted pairwise communication with controlled, coordinator-mediated coordination. Hub-and-spoke diagram of the Surgical Team Model: a central coordinator assigns tasks to and aggregates results from specialized agents such as architect, engineer, tester, and documenter.
Task
Roverall ( ↑ )
Ncalls ( ↓ )
Leff ( ↑ )
Ttotal ( ↓ )
Texec ( ↓ )
Base.
STM
Base.
STM
Base.
STM
Base.
STM
Base.
STM
Birthday Reminder
65%
86%
15
10
174
561
2273
8409
24.57s
128.1s
Equip. Borrowing
53%
78%
15
10
188
353
2192
7348
29.29s
131.1s
E-Commerce System
49%
78%
15
10
114
268
1664
8058
26.79s
172.4s
Inventory Mgmt.
43%
88%
15
10
155
302
1745
6321
21.69s
114.3s
Commodity Inv.
43%
88%
15
10
83
447
1607
8049
20.35s
134.0s
Table 1. Comprehensive Benchmark: STM Performance VS. MetaGPT
Task
Roverall ( ↑ )
Ncalls ( ↓ )
Leff ( ↑ )
Ttotal ( ↓ )
Texec ( ↓ )
Base.
STM
Base.
STM
Base.
STM
Base.
STM
Base.
STM
Birthday Reminder
71.0%
80.0%
7
7
473
463
6641
6396
126.0s
124.5s
Equip. Borrowing
92.5%
100.0%
7
7
558
626
7238
7553
136.7s
146.7s
E-Commerce System
87.1%
87.1%
8
8
645
656
8699
9260
161.2s
171.8s
Inventory Mgmt.
84.8%
87.2%
8
8
1070
861
1174
1174
225.3s
462.8s
Commodity Inv.
74.6%
74.3%
8
8
1078
878
12650
12802
149.3s
166.7s
Table 2. Comprehensive Benchmark: STM Performance VS. Claude Code
Appendix figures & tables1 asset
Supplementary material from the paper’s appendix.
Appendix
Figure 3. Cross-framework per-agent token consumption. Panels: (a) e-commerce, (b) movie-ticket booking, (c) commodity inventory, (d) personal accounting, (e) library management, (f) equipment borrowing, (g) birthday reminder, (h) volunteer registration. Each panel aggregates Django, Gin, and Express implementations of one task. The band is the mean ± standard deviation, and the marker indicates the smoothed optimum. One legend is shared across panels. Eight panels of cross-framework per-agent token curves with shaded standard-deviation bands and a marker at the smoothed optimum for each Web application task.
Today's agentic systems are typically designed, built, and deployed by single developers with full control over all participating agents. While this simplifies development, it limits the potential scale of agentic systems. We study multi-developer agentic systems (MDAs) which enable developers to collaboratively build agentic systems to push frontier capability beyond what any individual developer could achieve alone. We construct empirical and theoretical models of developers to compare MDAs with single-developer agentic systems (SDAs). On GAIA, a general assistant benchmark, a population of specialized developers whose individual agents do not exceed 35% accuracy reaches 70% when their agents are combined in an MDA. This MDA performs 12 percentage points above the median general-purpose SDA built with several times each specialist's development budget. Our theoretical models show how MDAs can incentivize specialization and encourage more efficient use of aggregated developer resources. Put together, our results make a preliminary case for MDAs as a new scaling axis by enabling developers with diverse capabilities and resources to advance agentic AI collaboratively.
Austin W. Hanjie, Carlos E. Jimenez, Zirui Wang +1
A multi-agent system can reduce latency on complex tasks by executing work concurrently. Several pioneering harness frameworks support multi-agent systems. However, the scalability of current multi-agent harnesses is often constrained by a central orchestrator's capacity to allocate tasks and coordinate workers. To address this limitation, we introduce Agensh, a scalable self-organized multi-agent harness without a central orchestrator: concurrent workers execute a multi-agent cooperation loop, continuously gathering context, claiming and self-assigning sub-tasks, taking action and sharing findings, verifying results, and merging progress in an asynchronous manner. The loop is supported by the agentic organization infrastructure comprising three components: a shared workspace holds proposed, ongoing, and completed work; a message interface lets workers communicate; and shared context retains reusable findings and work intentions. To test the scalability of Agensh, we evaluate it on the five hardest ProgramBench tasks with GPT-5.6-sol (high). Scaling from 1 to 128 agents raises the mean final test-pass rate from 19.31% to 28.78%, an approximately 49% relative improvement. Larger organizations reach comparable test-pass rates earlier. On pandoc, scaling from 1 to 1,024 agents raises the final test-pass rate from 33.89% to 55.06%. Worker trajectories further show that different forms of self-organized cooperation gradually emerges and standardizes as the organization grows. These results reveal the number of agents as a new scaling dimension for multi-agent organizations to expand the frontier of general intelligence, offering a practical solution for complex tasks under hard latency constraints or time budgets.
Large language model (LLM)-based multi-agent systems (MAS) have become a promising paradigm for complex information-seeking and reasoning tasks by enabling collaborative problem solving among specialized agents. However, existing MAS frameworks tightly couple task reasoning with coordination operations, including task selection, role assignment, message routing, and context management. As interactions grow, using powerful LLMs for these bounded control decisions introduces substantial token overhead and latency, limiting the scalability of agentic Web services. In this paper, we investigate whether coordination can be decoupled from expensive reasoning without compromising collaborative performance. We propose S1-MAS, a token-efficient multi-agent framework based on System One-guided computational division of labor. S1-MAS assigns bounded coordination decisions to lightweight System One models while reserving open-ended reasoning for capable LLM workers. Specifically, a lightweight controller selects inspection conditions, chooses subsequent tasks, and determines termination, while a compact reader retrieves condition-relevant evidence from authorized sources to support these decisions. Through a decision-evidence loop, selected tasks dynamically determine worker roles and source access, enabling adaptive collaboration without task-specific training. Extensive experiments on seven diverse benchmarks demonstrate that S1-MAS achieves superior accuracy while substantially reducing the inference cost. Across individual comparisons with AgentVerse, DyLAN, and SelfOrg on seven benchmarks, S1-MAS reduces GPT-4o token consumption by 44.9%-97.2% and measured end-to-end latency by 37.8%-93.0%. These results highlight its potential for scalable and cost-effective agentic Web applications.
Zihan Zhou, Xinzhe Hu, Hanxu Yang +2
University of Electronic Science and Technology of China · Southwestern University of Finance and Economics