Agent Capsules: Quality-Gated Granularity Control for Multi-Agent LLM Pipelines

> Paper: Agent Capsules: Quality-Gated Granularity Control for Multi-Agent LLM Pipelines > Author: Aninda Ray > arXiv: 2605.00410 | 2026-04-29

---

1. The Cost Dilemma of "N Agents = N Calls"

Imagine a typical multi-agent pipeline:

• Agent 1: Understand user intent
• Agent 2: Retrieve information
• Agent 3: Reason and analyze
• Agent 4: Generate the answer

Traditional execution:

• One LLM call per agent
• 4 agents = 4 calls
• High cost, high latency

Problems with naive merging:

• Combining all 4 agents into a single call
• But this causes:
• Loss of tool information
• Prompt compression
• Quality degradation

The question: how do you save tokens while preserving quality?

---

2. Agent Capsules: Adaptive Granularity Control

The paper proposes Agent Capsules:

Core idea: > Treat multi-agent execution as an optimization problem — minimize the number of LLM calls subject to quality constraints.

Technical approach:

1. Quality gating

• Not blind merging
• Instead, evaluate: "How much quality would be lost by merging?"
• If the loss is acceptable → merge
• If the loss is too large → keep agents separate

2. Adaptive execution

• Decided dynamically at runtime
• Not the same granularity for every task
• Simple tasks → merge
• Complex tasks → keep separate

3. Quality constraints

• Users define an acceptable quality floor
• The system guarantees quality never drops below it
• Efficiency is maximized under the constraint

4. Capsulization

• Each "capsule" = one execution unit
• May contain one or multiple agents
• Dynamically composed

It's like the art of project management:

• Not every task needs a meeting
• Simple tasks: an email suffices
• Complex tasks: a dedicated meeting is needed
• Adjust communication granularity by task complexity

---

3. Why Adaptive Beats Fixed Strategies

Problems with fixed fine granularity:

High cost:

• One call per agent
• N agents = N calls
• Even simple tasks are "over-executed"

High latency:

• Serial calls
• Accumulating latency
• Poor user experience

Problems with fixed coarse granularity:

Poor quality:

• Simple merging causes information loss
• Chaotic tool calling
• Degraded reasoning quality

Agent Capsules' advantages:

Dynamic balance:

• Merge when merging makes sense
• Split when splitting makes sense
• Optimal granularity

Quality guarantees:

• Explicit quality constraints
• System adapts to satisfy them
• Never "sacrifices quality to save money"

Cost optimization:

• Minimizes calls subject to quality
• Cheaper than fixed fine-grained
• Better quality than fixed coarse-grained

---

4. A Feynman-Style Judgment: Good Systems Know When to Simplify and When to Complicate

As Feynman said:

> "Knowing what to simplify is wisdom."

In multi-agent systems:

> "Agent Capsules' wisdom: not every situation needs the full multi-agent process. Handle simple problems simply, complex problems thoroughly. Adaptive granularity is engineering wisdom, not lazy compromise."

This also embodies Occam's razor:

• Don't multiply entities beyond necessity
• Don't multiply calls beyond necessity
• But when necessary, don't be stingy

5. Takeaways

If you're building a multi-agent system, ask yourself:

1. "Does my agent pipeline always execute at a fixed granularity?" 2. "Can I adjust dynamically based on task complexity?" 3. "Do I have explicit quality constraints?" 4. "Can adaptive execution cut costs while preserving quality?"

Agent Capsules reminds us: the efficiency of a multi-agent system lies not only in agent design but also in execution strategy.

When a system can intelligently decide "when to merge, when to separate," it finds an elegant balance between cost and quality. In the future agent economy, the best pipeline isn't the most complex one — it's the one that knows best when to simplify and when to complicate.

In intelligent system design, flexibility beats rigidity.