Why We Stopped Using LLM Agents to Control LLM Agents (Deterministic Multi-Agent FSM)

Description and use

Minions.AI replaced an LLM orchestrator with a deterministic TypeScript FSM, eliminating non‑deterministic control and achieving a 99.2 % automated completion rate.

Problem

Multi‑agent pipelines that rely on an LLM orchestrator suffer from non‑deterministic control flow, leading to loop oscillation, context poisoning, and unpredictable convergence, causing failures in ~30% of runs.

Outcome

Deterministic FSM in TypeScript eliminated non‑deterministic control, bounded iterations, and isolated agent contexts, yielding high reliability and cost predictability.

Workflow

• Harvest signals
• Draft content
• Critic review
• Revise draft (max 2 revisions)
• Stage to CMS

Evidence

The deterministic FSM pipeline has processed hundreds of scheduled content runs for Minions.AI with a 99.2% automated completion rate and zero loop oscillations.

Source

Loop oscillation. The critic agent would reject a draft for lacking specific technical details. The writer agent would add details, but slightly change the tone. The critic agent would then reject the new draft for tone issues, causing the writer to revert the details.

Source