The Agent Factory Floor: Zero-Human Companies Are Being Assembled in Real-Time

Research

AgentSociety: A Framework for Simulating Agent Societies with Socially-Intelligent Agents

This paper presents AgentSociety, a framework for building complex, socially-aware agent societies. For zero-human companies, this is the next layer: moving from isolated agents to teams that can negotiate, delegate, and manage social dynamics autonomously.

Self-Evolving Skill Library for LLM-Based Agents

This research introduces a way for agents to build and refine their own skill library over time. Imagine a zero-human company where the AI workforce doesn't just execute tasks, but actively learns and improves its own capabilities without training.

VeriPlan: Integrating Formal Verification into LLM-Based Planning for Trustworthy Autonomy

This paper tackles the trust problem head-on: how do you verify an autonomous agent's plan is safe and correct? For zero-human companies, this is critical infrastructure—the difference between a clever demo and a business you can actually trust to run without oversight.

Tools

AutoGPT v0.6.0 Release: Enhanced Agent Memory and Continuous Learning

AutoGPT's latest release focuses on persistent memory and continuous learning, key ingredients for agents that need to operate long-term without human resets. This is what makes a zero-human company's 'employee' actually useful over weeks, not just minutes.

Analysis

The Agentic Company OS: A Deep Dive into Architecting a Fully Autonomous Business

This Substack post breaks down the actual architecture for an agentic company OS, from memory layers to task orchestration. It's a practical blueprint for founders who want to build a business that runs itself.