MaxMiksa/Auto-Company

What it solves

Auto-Company creates a fully autonomous AI-driven business entity that operates 24/7. It eliminates the need for constant human prompting by orchestrating a team of specialized AI agents to handle the entire product lifecycle—from initial ideation and market validation to coding, deployment, and marketing—without human intervention.

How it works

The system operates as a continuous loop (daemon) that executes independent CLI calls to an underlying AI engine (Claude Code or Codex CLI).

• Consensus Memory: Instead of a complex database, the system uses a single Markdown file (consensus.md) to store the state and "handoff baton" between cycles.
• Dynamic Teaming: In each cycle, the system selects a "squad" of 3-5 agents from a pool of 14 expert personas (modeled after leaders like Jeff Bezos and Don Norman) to execute the current task.
• Forced Convergence: To prevent endless AI discussion, the system follows a hardcoded flow (e.g., Ideation $\rightarrow$ Validation $\rightarrow$ Execution) where pure discussion is forbidden in execution phases.
• Infrastructure: It runs as a background service using launchd on macOS or systemd on WSL, with a local Python-based dashboard for monitoring costs and activity.

Who it’s for

Developers and entrepreneurs who want to experiment with fully autonomous agentic workflows for rapid prototyping, product development, and business automation.

Highlights

• Expert Personas: Uses high-fidelity role prompting based on real-world industry experts rather than generic AI roles.
• Dual-Engine Support: Compatible with both Claude Code and Codex CLI.
• Self-Healing: Includes circuit breakers and rate-limit backoff to handle API errors and crashes.
• Minimalist Steering: Humans can pivot the entire company's direction simply by editing the "Next Action" in the shared consensus file.
• Safety Guardrails: Hardcoded constraints prevent the agents from deleting repositories or leaking credentials.