Skip to content
ARK360

How it works

One person. One agent. Bounded authority. Full audit. Then multiply.

ARK360 does not automate processes. It amplifies roles. The unit is one person and one agent, inside a policy envelope, on rails the organisation owns. Prove one, then multiply.

A complete reference platform, now selecting its first operators.

The unit

What one amplified role looks like.

Five stages, in this order. The order is the argument: verification comes before delegation, and judgement returns to the person before anything is recorded.

  1. Role

    One person whose time disappears into coordination, preparation and follow-up.

  2. Verifiable workflow

    The work written down so its output can be checked. Verification precedes delegation.

  3. Bounded agent authority

    What the agent may do, what it must hand back, what it may never touch. Role-scoped, in writing, before any code.

  4. Human judgement

    The decision returns to the person at the point where judgement is the work.

  5. Auditable outcome

    What acted, under whose authority, with what result. A record, not a recollection.

Verification precedes delegation. Judgement returns to the person.

The surface

Console, agents, integrations.

The operator console is where roles, workflows, approvals and the audit trail live. An operator configures what each agent may do per role, and reads what it did afterwards.

Agents reach your systems through governed integrations: the property-management platform, access control, finance and messaging. Each integration is scoped to what the role needs and nothing wider.

Operator controls

The three questions a reviewer asks first.

Identity and access
Agents authenticate as themselves through your directory, not through a shared service account. Each holds the narrowest authority that does its job, and appears as a principal you can see and revoke.
Data residency and tenancy
The build runs in an Australian region on infrastructure you can point at. Your systems remain the record; ARK360 holds no parallel source of truth.
Policy rails and approval chains
What an agent may do, what it must hand back and what it may never touch are written before any code, in a sheet a non-technologist can read. Approval chains follow your delegations.

Autonomy

Autonomy earns its way in.

Every workflow moves up a ladder, and only when its verification arrives.

  1. Level 01Assist

    The agent prepares and suggests. The person does the work, and every suggestion is checkable against it.

  2. Level 02Agent in the loop

    The agent carries the task. The person approves what matters before anything leaves the room.

  3. Level 03Artefact deleted

    The intermediate artefact goes. The agent works from the system of record, and the record writes itself.

  4. Level 04Domain autonomous

    Inside one bounded domain, the agent acts under its delegated authority, with the hand-back conditions written first.

  5. Level 05The captain

    The person commands a set of agents across domains. Judgement stays human; the grind does not.

A workflow ascends the moment its verification arrives. Never before.

The doctrine

Eight rules that decide what gets built and what does not.

  1. Processes do not adopt AI. People do. Put the agent inside the role.

  2. Make work verifiable, then delegate it.

  3. Amplify the person. Instrument the workflow. Recompose from evidence.

  4. Automation takes tasks. People take command.

  5. Authority is delegated, bounded, auditable and role-scoped, or it is not delegated.

  6. Delete artefacts. Never digitise them.

  7. Bold outcomes require conservative scaffolding.

  8. Protect persons, not positions.

The measure

One question closes every build.

Did this make the person more capable? If the answer is no, it is parked, in writing, with the reason.

Apply for the Founding Pilot Programme