Understand this story
On October 25, 2026, Shao Meng shared a Claude blog post's view on when to fold sub-agents into the main loop when building Agents. The blog argues that on the Claude Platform, a single main loop plus on-demand loading of Skills/tools often yields better quality, lower cost, and lower latency across multiple internal comparisons; one sub-agent per team plus aggregate orchestration often replicates the organizational chart, bringing problems such as context loss, Frankenprompts, and localized evals, and when one customer compared a multi-agent harness with a single loop, the latter was about 2× faster. The blog suggests reserving sub-agents only for scenarios such as parallelism, context bloat, or conflicts between tools and instructions, and offers a migration path of first building end-to-end evals and folding one at a time.
AI-generated from public reports
Latest development · Report
Kimi quota consumption too fast, Anthropic advocates folding sub-agents back into the main loopDevelopments
1 confirmed developments
Reports and discussion timeline
Browse reports and discussion by time. Their evidence levels remain distinct.
Oct 27
TMTPost · AGI
Kimi quota consumption too fast, Anthropic advocates folding sub-agents back into the main loop
After the launch of K3.1 Pro, multiple users reported that quota consumption was too fast, and Moonshot AI announced in the early hours of October 26 that it would reduce unnecessary sub-agent calls and output length. TMTPost reviewed the viewpoints in Anthropic's article "Building Agents: When to Fold Sub-Agents Back into the Main Loop": multi-agent systems consume about 15 times the tokens of ordinary conversations and suffer from information decay and personality fragmentation, and loading sub-agents on demand as Skills can cut over 80% of system prompts. Anthropic has also open-sourced the single-agent-loop-kit tool, enabling models to execute tasks only in single-agent form.
TMTPost · AGIRelated discussion
TMTPost · AGI joined this discussion
This post contributes to heat; it is not a formal report or confirmed fact.
36Kr · AI
Anthropic proposes folding sub-agents back into the main loop, responding to the controversy over Kimi K3.1 Pro quota consumption
Anthropic published the article "Building Agents: When to Fold Sub-Agents Back into the Main Loop," stating that a single agent with skills often outperforms multiple sub-agents dividing labor in quality, cost, and latency, with far lower token consumption than the latter, and open-sourced the single-agent-loop-kit tool. The article notes that multi-agent systems consume about 15 times the tokens of ordinary conversations, and converting sub-agents into Skills can cut over 80% of system prompts. After Moonshot AI's K3.1 Pro launched on October 24, multiple users reported excessively fast quota consumption, and the company said in the early hours of October 26 that it would reduce unnecessary sub-agent calls and output length.
Phoenix Technology
Kimi K3.1 Pro quota depletes too quickly; Anthropic argues for folding sub-agents back into the main loop
After the launch of Moonshot AI's K3.1 Pro, multiple users reported that quotas were being consumed too quickly. The company apologized in the early hours of October 26 and compensated users with quota, while stating it would reduce unnecessary sub-agent calls and output length. The article cites Anthropic's piece "Building Agents: When to Fold Sub-Agents Back into the Main Loop," stating that token consumption in multi-agent systems is about 15 times that of ordinary conversations, and that converting sub-agents into Skills on demand can cut over 80% of system prompts. Anthropic has also open-sourced the single-agent-loop-kit tool, enabling models to execute tasks in a single-agent form only.
Oct 25
The Decoder
Anthropic says a single agent often outperforms multi-sub-agent architectures, and releases a migration toolkit
Anthropic recommends that enterprises evaluate whether to consolidate sub-agents, because a single agent that loads context on demand often outperforms multiple specialized sub-agents in quality, cost, and latency. Starting with Opus 4.5, as well as newer models such as Opus 5.5 and Fable 5.1, complex plans can be managed directly within a single loop; in one comparison, the single-agent architecture ran at roughly twice the speed of the multi-agent approach. Anthropic released a migration toolkit on GitHub and cautions against consolidating blindly, because in one customer-service agent test, merging in sub-agents reduced latency but broke the existing workflow.
Shao Meng
Claude Blog: When building Agents, when to fold sub-agents into the main loop; progressive disclosure of Skills is often better than multi-expert orchestration
The Claude Blog discusses when to fold sub-agents into the main loop when building Agents, arguing that on the Claude Platform, a single main loop plus on-demand loading of Skills/tools often yields better quality and lower cost and latency across multiple internal comparisons. The article notes that one sub-agent per team plus aggregate orchestration often replicates the organizational chart, bringing problems such as context loss, Frankenprompts, and localized evals; one customer, when comparing a multi-Agent harness with a single loop, found the latter about 2× faster. Sub-agents are recommended only for scenarios such as parallelism, context bloat, or conflicts between tools and instructions, and the article provides a migration path of first building end-to-end evals and folding in one at a time.
Heat trend for this story
- Current
- 15
- Period peak
- 26 Oct 27 20:00
- 24-hour change
- +31%
Move the pointer or select the chart to inspect hourly heat. Use the left and right arrow keys with a keyboard.