Orchestration research
Deterministic AI agent scheduling selects work from durable project state using explicit readiness, dependency, ownership and capacity rules. Given the same authoritative state, the scheduler should reach the same allocation decision. Models may help plan tasks, but they should not improvise the control-plane transition that grants execution authority.
A work slice is ready only when prerequisites, required decisions and capacity constraints are satisfied. Those facts should be queryable and recorded, not reconstructed from conversational context.
Deterministic rules make allocation explainable and safe to retry after a process restart.
AI can help decompose a plan or estimate which specialist role a task needs. Once accepted, the control plane turns that plan into explicit state.
Letting a model decide the final ownership transition at runtime makes duplicate allocation and recovery harder to reason about.
Nexus Engine reads PostgreSQL-backed authority, issues fenced assignments and records every transition. Capacity is governed across active projects and worker roles.
Scheduling optimises safe, verified progress rather than maximum concurrent prompts.
Nexus launch access
Register your interest now and we will let you know when the autonomous AI software engineering control plane is ready.
