Skip to main content
HUMΛN
HUMΛN
←Community
Architecture
Architecture

What is Human-AI Orchestration (HAIO)? — practitioner reference

HUMΛN Team··14 min·Engineers & evaluators

You already read the launch narrative. This page is the reference sheet you hand an evaluator: what HAIO names, what must be true in the protocol, and what is not implied by “agent framework” or “model router.”

Definition (protocol-level)

Human-AI Orchestration (HAIO) is a class of systems where work crosses human and machine actors under explicit delegation, policy, and provenance — so that who acted, under what authority, and with what uncertainty can be verified at the handoff, not reconstructed from logs after the fact.

That handoff is the differentiator: not the model, not the DAG shape — accountability at the boundary where judgment, scope, and irreversibility meet.

Why this is a category, not a sticker

Adjacent stacks optimize for different primary goods:

Adjacent stack Primary optimization
Throughput-first connectors / RPA Connectivity and volume
Agent frameworks Tool routing and developer velocity
Identity providers Authentication to a tenant
LLM call routers Model selection, retries, spend

HAIO optimizes for governed coordination across those layers: portable identity, capability truth for routing, policy and escalation, and machine-attributable provenance — so the same story holds in audit prep, engineering postmortems, and operator review.

HUMΛN is not “an agent marketplace company” and not a staffing platform. Marketplace and Workforce are surfaces of the trust layer between people and AI — they are not the category claim.

What changes when you take HAIO seriously

  1. Handoffs are first-class — Every meaningful transition emits or preserves a decision-shaped artifact (receipt), not only a log line. See What a decision receipt looks like.
  2. Delegation precedes execution — Agents do not inherit user power by UI session alone; scopes are explicit and revocable (Passport + HumanOS).
  3. Uncertainty is routed — Low confidence triggers escalation, not silent best-effort completion (Fourth Law patterns in production governance).

HumanOS still enforces policies you define. HAIO does not mean the runtime invents corporate judgment.

Practitioner checklist

Before you recommend a stack, ask:

  1. Can we name the actor (human, agent, service) for each externally visible step?
  2. Can we show the delegation that permitted the action?
  3. Can we cite the policy version that applied?
  4. If confidence was insufficient, what escalation path fired — and what record exists?

If those four are muddy, you are not debating HAIO implementation details — you are missing architecture.

Scroll-stopper: The handoff is the product. Everything else is plumbing until the handoff is attributable.

So that evaluators stop grading demos

Use this sheet when a vendor shows a polished agent and claims “orchestration.” Demand the four answers above. If they answer with a transcript, you are still buying literature.

Go deeper

Code & Docs

ctx.provenance

Agent SDK reference