Investigations

StateLoop

TypeScript February 2026

A multi-agent AI negotiation system where autonomous agents with distinct strategies compete through structured rounds.

StateLoop explores what happens when you give AI agents distinct personalities and let them negotiate. Each agent has its own strategy — cooperative, competitive, analytical, or deceptive — and they go through structured negotiation rounds trying to reach agreements.

The system manages state transitions as agents move through proposal, counter-proposal, concession, and agreement phases. Each round captures the full conversation history, strategy adjustments, and outcome metrics. You can watch agents adapt their approach based on what's working.

What made this technically interesting is the state machine architecture. Each negotiation follows a defined flow with clear transition rules, but the actual content and strategy of each move is determined by the AI agents. It's deterministic structure with non-deterministic behaviour inside it.

The dashboard shows real-time negotiation progress, agent strategy distributions, outcome statistics, and conversation logs. You can configure agent personalities, negotiation parameters, and scoring criteria.

What this project demonstrates

Multi-agent system with distinct negotiation strategies
State machine architecture for structured negotiation rounds
Real-time dashboard with progress and strategy visualisation
Configurable agent personalities and scoring criteria
Full conversation history and outcome metrics
Deterministic flow with non-deterministic AI behaviour

Technologies

AIMulti-AgentState MachineNegotiationDashboard

© 2026 Jonathan Leahy · v0.8.1-31-g196fa14