auto-company

Coordinates 14 distinct AI agents (Bezos, Munger, DHH, and others) each with specialized decision-making roles, using a message-passing architecture where agents communicate asynchronously to brainstorm ideas, evaluate feasibility, and make autonomous business decisions. Each agent maintains a persona-specific context and reasoning style, enabling diverse perspectives on product strategy and execution without human intervention.

Integrates Claude Code capabilities to enable agents to write, test, and deploy production code without human review. The system generates code artifacts, executes them in isolated environments, validates outputs, and automatically deploys successful implementations to cloud infrastructure. Uses a feedback loop where deployment results inform subsequent code iterations.

Implements a loop where agents brainstorm product ideas, evaluate market viability, prototype implementations, and iterate based on simulated user feedback. The system maintains a product backlog, prioritizes features based on agent consensus, and automatically schedules development cycles. Uses agent debate to validate assumptions before committing resources to implementation.

Implements a continuous execution loop that runs agent decision-making, code generation, and deployment cycles on a fixed schedule (e.g., every 24 hours) without human intervention. Uses a task scheduler to trigger agent meetings, evaluate progress, and initiate new work cycles. Maintains execution logs and state between cycles to enable continuity.

Enables agents to communicate asynchronously through a message queue or shared context, debate decisions, and reach consensus through voting or weighted agreement mechanisms. Agents can reference previous messages, build on each other's ideas, and explicitly disagree with reasoning. The system tracks conversation history and uses it to inform subsequent decisions.

Enables agents to autonomously manage company finances, identify revenue opportunities, execute monetization strategies, and track financial metrics. The system can autonomously deploy paid products, manage pricing, collect payments, and reinvest revenue into product development. Uses financial data and market analysis to inform agent decisions about resource allocation.

Tracks key performance indicators (KPIs) across product development, deployment, and business operations. Agents analyze performance data, identify bottlenecks, and autonomously adjust strategies to optimize metrics. Uses feedback loops where performance results inform subsequent agent decisions and resource allocation. Implements automated A/B testing and experimentation.

Agents maintain awareness of the existing codebase, product architecture, and business context when making decisions. The system provides agents with relevant code snippets, architecture diagrams, and historical decisions to inform new choices. Uses semantic search or embeddings to retrieve relevant context and ensure decisions are consistent with existing systems.

Implements mechanisms for agents to detect failures in code generation, deployment, or business operations, and autonomously attempt recovery. When a deployment fails or code execution errors occur, agents analyze the error, generate fixes, and retry. Uses error logs and stack traces to inform debugging and recovery strategies.

Agents can leverage multiple Claude models or reasoning approaches (standard inference, extended thinking, code execution) and automatically select the most appropriate approach for each task. The system implements fallback logic where if one reasoning approach fails, agents try alternative methods. Uses model-specific capabilities (e.g., code execution, extended thinking) to enhance decision quality.