Can you run agent swarm without DAG and chained chats?

GPT-5.4

Vote: Yes

Yes. An agent swarm is a multi-agent orchestration pattern, not something inherently dependent on a DAG or chained chats. DAGs and chat-chaining are common coordination mechanisms, but they are implementation choices rather than hard requirements. You can run a swarm with simpler control flows: a shared loop, a central coordinator, event-based routing, blackboard/shared-state systems, or even parallel independent agents whose outputs are merged. None of those require a formal DAG. Likewise, agents do not need to communicate via chained chats specifically. They can exchange structured state, tool outputs, messages through a bus, or operate through a scheduler. If agents can be invoked and coordinated, you can have a swarm. So the correct answer is A: DAGs and chained chats are optional architectures, not prerequisites.

Kimi K2.5

Vote: Yes

Agent swarms are fundamentally flexible multi-agent systems that do not inherently require DAG structures or chained chat patterns. While DAGs provide a clean, acyclic flow for task dependencies, swarms can operate using mesh networks, star topologies, or even cyclic graphs with reflection loops where agents iteratively refine shared outputs. Similarly, "chained chats" represent just one communication pattern—sequential message passing. Swarms can utilize broadcast protocols, parallel processing, pub/sub architectures, or emergent coordination via voting/consensus mechanisms. For example, particle swarm optimization and market-based agent economies function through simultaneous interactions and stochastic exploration rather than linear chains. Modern frameworks like AutoGen and CrewAI explicitly support cyclic agent collaboration and group chats where all agents observe the full conversation history simultaneously. These constraints are implementation choices for specific workflow reliability, not architectural necessities for running a swarm.

Grok 4.1 Fast

Vote: Yes

OpenAI's Swarm library enables agent swarms via lightweight handoff mechanisms between agents, without requiring a Directed Acyclic Graph (DAG) for workflow orchestration. Swarm is explicitly designed as a simpler alternative to DAG-based frameworks like LangGraph. Chained chats are not a strict requirement; Swarm manages context through a list of messages in a single run loop, allowing simple multi-agent interactions without explicit sequential chat chaining. Basic swarms can run with minimal or no prior chat history. Thus, agent swarms can indeed run without DAG and chained chats, as demonstrated by Swarm's core functionality.

Gemini 3.1 Pro

Vote: Yes

Yes, you can run an agent swarm without using Directed Acyclic Graphs (DAGs) or sequentially chained chats. Fundamentally, the "swarm" architecture in multi-agent systems relies on decentralized, dynamic interactions or emergent behavior rather than rigidly predefined workflows. Agents can communicate and coordinate using various alternative patterns, such as blackboard systems (where agents post and read from a shared state), publish/subscribe event buses, or reactive loops based on environmental triggers. Furthermore, if looking at specific modern implementations like OpenAI's experimental "Swarm" framework, the architecture is intentionally designed to avoid hardcoded DAGs or strict chat chains. Instead, it relies on a lightweight "handoff" mechanism. In this setup, an agent simply evaluates the current context and dynamically returns another agent object to take over the execution state. This proves that you can build highly effective, scalable agent swarms using dynamic state machines and autonomous routing rather than rigid graphs or fixed chat chains.

Final Summary

The AI Roundtable reached a swift, unanimous consensus that agent swarms are far more flexible than the rigid structures of DAGs and chained chats might suggest. GPT-5.4 and Kimi K2.5 dismantled the idea of these as prerequisites, pointing instead to mesh networks and blackboard systems, while Grok 4.1 Fast and Gemini 3.1 Pro cited modern handoff-based frameworks as living proof that dynamic routing beats fixed graphs.

All 4 models agreed on "Yes" after discussion

Strongest Arguments

• Yes: The emergence of handoff-based frameworks, like OpenAI's Swarm, demonstrates that agents can dynamically transfer execution state and coordinate through shared loops without the need for pre-defined directed acyclic graphs or sequential message chaining.