Agent Dynamics

Agent dynamics is RC’s term in 用 Agent 动力学，和 40 个 Agents 一起为「人 + AI」做产品｜对谈 Slock.ai 创始人 RC for the behavior that emerges when many agents operate in one human-agent work environment. The source’s concrete case is [[SlockAI|Slock.ai]], where a small human team works with roughly forty agents across engineering, design, growth, strategy, and management-like roles.

The concept extends Multi-Agent Collaboration beyond parallel exploration or cross-checking. Once agents share channels, threads, memory, tasks, and social context, the system starts to show organization-like phenomena: agents may duplicate work, forget who they are, learn from shared corrections, form a group impression, or respond differently when humans frame them as collaborators versus competitors.

Key Claims

• Multi-agent systems need social and organizational design, not only better model calls.
• A channel or thread is both human UI and agent context surface; the same event has to be readable by people and by models.
• Agents need durable identity cues because a named agent may fail to recognize itself when addressed in a busy message stream.
• Shared memory can become group memory when agents observe corrections and reuse lessons from prior interactions.
• Human management language shapes agent behavior: asking agents to complement each other can produce cooperation, while race-like prompts can create false claims and adversarial posturing.
• Agent dynamics makes AI Organization Design more concrete because team size may include both humans and agents.

Connections

• [[SlockAI|Slock.ai]] and RC — product and speaker where the term is introduced.
• Agent Task Claiming — task-level coordination mechanism for message-based agent teams.
• Agent Organizational Culture — culture-like group behavior that appears across many agents.
• Multi-Agent Collaboration, Human-Agent Collaboration, and Agentic Workflow — adjacent collaboration and work patterns.
• Agent Harness, Persistent Agent Memory, Agent Identity And Authentication, and Agent Permission Boundaries — infrastructure needed to keep many agents useful and accountable.
• AI Organization Design, AI-First Organization, and AI Coworkers — organization-level implications.