Agentic Logic
Agents are where the system stops recording and starts doing. Each one serves a domain — sales, purchasing, finance, service — and works within explicit limits: what it may read, what it may change, up to what value, and what must wait for a person.
- Domain agents with written autonomy limits
- An agent proposes, a person decides on critical steps
- Process templates instead of hand-built workflows
- Every action logged: who, what, why and on what evidence
From recording to acting
Business software has always held an accurate account of what already happened, kept current by people typing, and preparing the next step was left to them. This layer is the difference: the job closes, so the invoice is drafted; stock crosses a threshold, so the replenishment is prepared; a delivery slips, so the affected orders and customers are listed with a proposed answer.
The unit of work is the job, not the screen it happens to live on.
Limits, written down
Autonomy is only useful when someone owns its limits. So every agent has a scope that a person can read: the records it may touch, the actions it may take on its own, the value up to which it may act, and the actions it may only propose. Some operations are reserved for people under every configuration: approving a payment, signing a commitment, changing who may approve what.
Limits are set per process as well as per agent, because the same agent may reorder office supplies on its own while a reorder of a critical component goes to a person.
How a proposal travels
An agent asks the Super Intelligence layer above for reasoning — read this offer, compare these three, draft this answer — and gets a result back. It then acts only in terms the Ontology below defines: the objects that exist, the actions allowed on them and the rules they must satisfy. Anything beyond its own limits becomes a proposal in the approval inbox, with the sources it used and the risk it sees. A person approves, changes or rejects it. The Data layer stores the outcome, and Security and Sovereignty checks the permissions and records who did what.
Building the system itself
The same path applies to Hurukan’s own development. A new field, a new approval step or a new report is described, an agent writes it, the tests and review run, and a person decides whether it ships. Faster gates, not fewer of them.
What the layer does
Domain agents
Each agent belongs to a domain and a role, and sees only what its role may see: purchasing data for a purchasing agent, service data for a service agent.
Autonomy limits
Set per agent and per process: which actions run on their own, which become proposals, and which are reserved for people.
Approvals
Critical steps — money leaving, a customer contacted, a commitment made, a change to the system — are proposed with their sources and their risk, and wait for a person in one approval inbox.
Process templates
Recurring work such as a quote, an inbound invoice or a complaint follows a template of stages, so the agent and the people involved know what comes next and who owns it.
Agents that extend the system
A change to the system is described in plain language. An agent writes it against the ontology, the tests run, a person approves it and the release is recorded.
Action log
Every action by a person, an agent or the system is written down with what was observed, what was concluded and who approved it, readable months later.
Works with
Decisions and Tasks
One approval inbox for agent proposals with sources and risk, and tasks that always have an owner.
Commerce
Opportunities, quotes in variants, versioned offers, contracts, orders and billing schedules on one directory.
Procurement
Requisitions, a supplier register with risk status, RFQs, offer comparison, approved purchase orders, deliveries and returns.
Finance
Finance and controlling: budgets from plan to forecast, inbound invoices, invoicing, payments and cash flow, linked to accounting.
Questions
Who approves a payment?
A person. Agents prepare, check and propose; a critical operation such as a payment is approved by a person, and the security layer enforces that rule on the server, independently of the agent's instructions.
Who sets an agent's limits?
The people responsible for that area, per agent and per process. Changing a limit is itself a change to the system, so it goes through the same approval and audit as any other.
What happens when an agent is wrong?
A wrong proposal costs the time it takes to reject it, and the rejection is recorded so the pattern can be corrected. Where an action can be undone it carries its reversal; where it cannot, it is one that waits for a person in the first place.
Build the company's operating system with us
Hurukan is pre-launch. If you run a company and want a sovereign, agentic system rather than another suite, we would like to hear what you need.