Skip to main content
HUMΛN
HUMΛN
←Community
Vision
Vision

Introducing HAIO — the category for trust between people and AI

HUMΛN Team··12 min·Technical + General

Friday night: the agent books meetings, files tickets, and pings Slack. The README says production-ready. The demo reel looks expensive.

Monday morning, security asks a question no README answers: who is allowed to do that on behalf of which human — and how do we prove it to an auditor six months later?

That is not a missing middleware plugin. It is a category gap. We named it HAIO — Human-AI Orchestration so buyers, builders, and regulators can stop arguing about models long enough to argue about trust architecture.

What most “orchestration” sells (and quietly skips)

Vendors say orchestration and mean routing model calls (and maybe tools). The human is optional. The audit trail is a pasted chat transcript. When something emails a customer, that transcript becomes literature — not evidence.

HAIO means the opposite build order:

  1. Identity — humans and agents as principals (Passport / DIDs), not anonymous API keys with a persona.
  2. Delegation — scopes that narrow and expire; no permanent superpowers because the UI session felt friendly.
  3. Policy — HumanOS enforces the rules you define; it does not invent company judgment for you.
  4. Provenance — decision receipts by default: who, under which grant, which policy version, what happened.

Model choice is still important. It is just not the load-bearing beam.

One founder story we kept hearing

A team ships an internal agent that “helps with ops.” It works in the happy path. Then legal asks for the chain: Passport → delegation → tool → receipt. Nobody has it. The postmortem is not “temperature was too high.” It is “we built a remote shell with better JSON.”

HUMΛN exists so that story ends differently — same Friday velocity, Monday-grade accountability. Passport, Capability Graph, HumanOS, Workforce Cloud, and Companion are the surfaces; HAIO is the name of the category they inhabit.

Five principles (in one breath)

  1. Human sovereignty — identity is not our database row; portability and consent are structural.
  2. Capability truth — skills grounded in demonstrated evidence, not keyword cosmetics.
  3. Guidance, not control — AI proposes; humans approve irreversible or high-risk acts.
  4. Provenance by default — receipts, not vibes.
  5. Sovereignty again (on purpose) — your data is not our moat; trust is.

We repeat sovereignty because markets will keep offering “free” conveniences that centralize. HAIO is the refusal to build another silo with a chat UI.

So that the conversation stops mapping you to cron-for-prompts

If we only say “orchestration,” buyers map us to prompt routers. If we say HAIO, we force the conversation onto Passport, Capability Graph, HumanOS, and Workforce — infrastructure, not “which model.”

Categories are not vanity. They are shared vocabulary for people who already feel the pain and lacked the name.

Scroll-stopper: If you cannot name the delegation on an agent action, you do not have orchestration — you have a ghost with API keys.

Builders: start with the receipt

Before you add another tool call, write the one-paragraph incident narrative you could defend:

  • Actor — human vs agent vs service
  • Delegation — scopes, expiry
  • Policy version that applied
  • Outcome and escalation if confidence was low

If you cannot produce that paragraph, you are not blocked on temperature — you are blocked on architecture.

Go deeper

Code & Docs

ctx.provenance

Agent SDK reference