Modelling Agent Systems with Erlang (2004) [pdf] (erlang.org)

This 2004 paper presents a practical model for building BDI (Belief-Desire-Intention) multi‑agent systems using Erlang and OTP, arguing the language’s lightweight processes, message-passing, and built‑in fault‑tolerance make it a strong platform for distributed, concurrent agents. The authors map classic BDI concepts into Erlang artifacts (beliefs/goals as tuples, plans as records with trigger/precondition/postcondition/invariant/body), and describe an agent architecture composed of cooperating processes (event receptor, state server, executor and main process). They demonstrate features for plan exchange, dynamic reconfiguration and fault‑tolerance, positioning Erlang as a way to get high availability and resilient coordination in multi‑agent deployments. Technically, plans are closures evaluated against executor state, with invariant checks after each step and explicit cancellation on failure; bodies are sequences of {action, …} or {goal, …} tuples so actions can return values stored in the agent state. The paper includes Erlang snippets (gen_server-style handlers) showing event handling, belief updates, action/goal production and plan execution loops. For the AI/ML community this provides a concrete, scalable pattern for implementing distributed reasoning, coordination and reactive planning—useful for multi‑agent RL, robotics, and large-scale distributed AI services where concurrency and robustness are essential.