AI-native operations: design the work around the result, not the task
Why faster steps leave the old organisation in place, how to plan work from the result down, and the four numbers that decide how far to trust the system.
No status reports to read. One decision to make.
A manager gets a request, and AI helps sort it. The manager prepares an offer, and AI helps write it. The manager checks a status, and AI writes a short summary. The director needs a report, and AI formats it.
Every step is faster now. Yet the manager still carries the context from one system to the next, starts every action, watches whether it happened and puts the whole picture together by hand. The steps were sped up. The way the work is organised stayed the same.
That is an AI-assisted company. An AI-native company starts somewhere else: with the question of how the work should run at all, once a machine can do a large share of the thinking.
Check where your company is today
Five questions about one process you know well, such as handling a customer order from request to delivery. It takes two minutes.
Is this process AI-assisted or AI-native?
Think of one process that runs every day: an order, a supplier payment, a customer request.
Your answers stay in this browser. We count only which button was pressed, never anything you type.
Most processes score one or two. That is the normal starting point, and it shows where to build first.
We automate the wrong thing
The usual approach takes an existing process, finds one step that AI can speed up and adds a model there. Then the next step, and the next. Each tool helps, and none of them changes who holds the work together. That job stays with people: remembering what was promised, noticing what is late, deciding what happens next.
In an AI-native model the roles swap. People set the goals, the rules and the limits. A managing layer keeps the context, runs the connected actions, watches for anything going off course and brings a person in when a decision is needed.
Who carries the work between systems
Plan from the result, not from the tasks
Most companies are organised around tasks: call the supplier, check the payment, confirm the date, update the CRM, prepare the report. The owner does not need any of those actions. The owner needs a result, for example: the goods reached the customer on time, at the agreed margin and without unacceptable risk.
An AI-native process is designed from that result down. For each case the system works out what is already known, what is missing, which actions are allowed and which decisions need a person. Tasks do not disappear. They follow from the result, and they stop being the way the business is organised. People stop being the routers who must pass every step along by hand.
The loop every case runs
Example: one order at a trading company
A request for a delivery comes in. In the usual setup a manager starts a chain of actions: checks the customer, asks suppliers for terms, compares the offers, talks to the freight team, watches the payments, updates the statuses and tells the owner when something goes wrong.
In an AI-native setup the request gets its own digital case manager: an agent responsible for moving it forward. It keeps the state of the case and works out the next allowed step, without going beyond the rights it was given.
One order, three kinds of workers
The owner should not receive a list of tasks and overdue items. The owner should receive a decision, already framed:
The difference is that nobody has to keep asking the system what is happening. The system brings the situations that need a decision to the person who must make it.
This works only under three conditions: the data is confirmed, the AI’s rights are limited, and the system can say “I am not sure”. Without them, the result is not control. It is a convincing imitation of control.
Five parts of an AI-native platform
Each of the five parts answers a question that, in an AI-assisted company, a person answers from memory every day. If you took the test above, the parts your process is missing are marked.
What an AI-native platform is made of
What must be achieved?
The result for each kind of case, with its limits: the deadline, the cost, the quality and the risk that is acceptable. Without it the system can only follow tasks, and nobody can tell whether a case went well.
An order is done when the goods are with the customer on the agreed date, the margin is within the agreed range and the payment is matched.
What do we know about this case?
The things the business deals with and how they connect: customers, suppliers, orders, contracts, payments. Plus the history of decisions, the documents, the current state and where each fact came from.
The case for one order links the customer, the contract, three supplier offers, the freight quote and the payment, each with its source.
What happens next?
The mechanism that starts agents and ordinary software, tracks what depends on what and moves each case to its next step. This is the part that replaces the manager as the person who passes work along.
When the freight quote arrives, the margin is recalculated and, if it falls below the limit, the case goes to the owner.
What may the system do alone?
What AI may do by itself, when it must ask for confirmation and when the process must stop. Money, law, signatures and anything sent outside the organisation stay with people.
The agent may compare offers and prepare the choice. A person approves the supplier and the extra freight cost.
How do we know it worked?
How the system proves a result was reached, spots a mistake, counts what each case cost and improves the process over time. Without this part nobody can decide when to give the system more freedom.
Every finished order shows the time it took, what it cost, whether a person had to step in and what went wrong.
\1
Four layers of an AI-native company
There is no need to replace those systems at once. That path is expensive and risky. It works better to build a managed layer of cases on top of them and widen its freedom step by step.
Read also · 11 minHow an AI agent earns permissions: read, advise, draft, record, executeThe fourth part in detail: five levels of permission and how an agent climbs them.
The organisation changes too
If the system can carry a large share of routine operations, a harder question follows: why keep an organisation built around passing information by hand?
Some roles exist today because someone has to carry information between departments, check statuses, remind people of what they promised, collect reports and notice when things do not match. Over time a manager’s job moves from coordinating by hand to looking after a portfolio of results, handling the exceptions and improving the rules the work runs by.
Cutting staff should not be the goal. The goal is to let the business handle more without coordination costs growing at the same pace.
The trap: autonomy for its own sake
It is possible to build an impressive system in which dozens of agents talk to each other all day and the business gains nothing. So an AI-native process is judged neither by the number of agents nor by how independent they are. It is judged by four numbers.
\1
Five steps of freedom, and a ceiling
Your plan for this week
Steps for this week
Ticks stay in this browser.
Choosing agents, frameworks and tools comes later. The first piece of architecture is the operating model: people, agents and ordinary software as three kinds of workers, each doing what it does best. A company used to be an organisation of people who use software. An AI-native company becomes a system for reaching its goals, in which people, agents and software work together.
Pick a question and we will answer it here.
Not on the list? Ask the team your question
Was this article useful?
Thank you. What should we write about next?
More from the blog
- 5 things to try
AgentsWhat an agent harness is, and why it matters more than the modelThe same AI model can sit near the bottom of a benchmark or near the top. The difference is the harness, the code around the model that runs the loop, hands it tools, decides what it sees, keeps its memory and controls what it may do. What that means for agents that handle a company's invoices, orders and requests.
Denis Zaitsev · · 14 min - 5 things to try▶ Listen · 9 min
Getting readyHow to prepare your organisation for AI agentsBefore an AI agent can do useful work, it needs a frame to stand on. Five parts of that frame, how to check which you already have, and why the first step is one process.
Denis Zaitsev · · 10 min