InstructionsRole, goal and limits, in plain words
Whose desk you serve and what result is expected.
Which steps and checks to do, in what order.
What you never do: money, law, signatures.
When to stop and hand the case back to a person.
A management layer on top of your data
Your documents, systems and conversations become one working model. Agents prepare the work. People decide.
1 / 8
Raw material
Documents, chats, meeting recordings, exports from old systems. Nothing needs tidying first: it all goes into one base, each piece linked to its source.
2 / 8
Leadership answers, the model fills in
In a facilitated session the leadership answers one question at a time. Where they disagree, they settle it in the room. Each answer is recorded with who gave it and fills a gap in the model.
3 / 8
A dictionary of the business
Fragments become objects: customer, contract, shipment, payment. One concept, one name, linked to everything it touches.
4 / 8
Laid out in databases
Every object gets its fields and its home. Records from different systems meet in one model.
5 / 8
Processes appear
The links show how work really moves: steps, desks, checks and branches, modelled on the process engine.
6 / 8
The engine breaks it first
Before anyone works by the model, the engine tries to break it: skipped steps, empty fields, missed deadlines. Every gap gets a rule.
7 / 8
Places for agents
Each step gets an executor: the system, an agent with a person, or a person alone. Money, law and signatures stay with people.
8 / 8
A desk for every role
The model becomes a desk for each role, with each task set the way that person works. Facts flow back and the model gets its next version.
Built in Dubai, for the UAE's ambitions
The UAE set the course. We build for it.
An AI-native government, an economy driven by AI, the best country in the world by 2071. Few nations aim this high, and say it this clearly. We share the ambition, and build the operating layer that turns it into everyday work.
- 2027Abu Dhabi Government Digital StrategyThe world’s first fully AI-native governmentThe model and desks for an AI-native agency
- 2031National Strategy for Artificial IntelligenceAmong the world’s leading nations in AIAgents on every desk. People decide.
- 2033Dubai Economic Agenda D33Double the economy, AED 100 billion a year from digitalDigital operations for companies that scale
- 2071UAE CentennialThe best country in the worldInstitutions that keep improving, fact by fact
The result: a desk for every role
Every desk has the same three parts. The agent has already checked the case and named its sources. The decision is yours: try it, approve the payment.
- QueueEvery task gets a score from its deadline, amount, risk and waiting time. The most important one is always on top.
- Case solverThe agent opens the case, calls other agents for each check and prepares the decision with sources. You press one button.
- DossierEverything that matters about the case in one place: the other party, contract, documents, history and who decides.
Treasurer’s deskZarvane Trading Group
From the procurement desk
Supplier payment: steel coils, lot 2 of 3
AED 412,80018:00
Prepared by the treasurer’s copilot
Ready to pay. All four checks passed.
- Invoice matches the contract and the delivery note
Contract C-118 · delivery note 4471 - Price within the approved band
Price band, Q4 steel - Counterparty check unchanged since March
Security desk - Budget line has room
Raw materials, Q4
Your decision
Harness, the organisation’s memory
Every source, decision and rule lands in one memory, with its origin. Six maps read it. Agents and desks work from it.
- Meetings
- Documents
- Decisions
- Email and chats
- Your systems
- Treasurer’s copilot
- Sales copilot
- Contracts expert
- Treasurer’s desk
- Minister’s office desk
Management map. Who answers for what, and how a goal reaches the people who do the work.
Agents work like new colleagues
Each agent has a job, an owner and rights. It earns more rights the way a new hire does, by a record of good work. Money, law and signatures always stay with people.
SkillsKnow-how written once, shared by every agent that needs it
- Match documents
- Check against a rule
- Extract data from a file
- Compare offers
- Draft a reply
Connectors (MCP)Standard links to the systems where your data lives
- CRM
- Accounting
- Email and chats
- Documents
- Bank statements
- Public registers
Agentworks from Harness: processes, decisions and rules
- Read
- Advise
- Draft
- Record
- Execute
On a person’s deskA prepared case
Everything checked, every fact with its source, the next step drafted.
- Checks
- Sources
- Draft
- Next step
A team of agents, not one bot
Every desk gets its own copilot. Experts serve several desks, strategists watch each unit, and a few agents work across the whole organisation. All of them work from the same memory.
Rights are earned, not given
An agent starts in the shadow. After a month of mostly “yes” and the owner’s approval it moves up one step.
When it cannot, it hands back
The case returns to the step owner with what the agent tried, on which data, and why it stopped.
Long jobs wait in their own queue
The case moves to “At the agent” with an expected time and comes back done, with a question, or failed.
One console for all agents
Status, spend and a full log of every action, the same as for a person. Any agent can be paused in one click.
Recorded with your name and time
You answer on your own desk
Yes, no, later or a question. A task is created only after a person says yes.
Red zone Money, law, signatures, anything sent outside, and changes to the rules. Here an agent only prepares. A person decides.
Modules
Nine ready modules on one model, and room for yours. A case moves between them without being retyped.
Deals
From request to quote, contract and invoice.
Tasks
Every task tied to a case. An agent can be the assignee.
Suppliers
One register, side-by-side scores.
Procurement
Brief, offers, comparison, order.
Invoices
From draft to closing documents.
Contracts and documents
Templates, two-step review, signature.
Counterparty checks
Monitoring, stop list, price control before payment.
Request intake
Reads the inbox, filters noise, opens cases on the right desk.
Leadership console
Instructions broken into tasks and followed up.
Every module is a directory with an engine
A module is a directory of your data, an engine that runs the rules on it, and a desk where the work is done. So a module for your own process is not a new product: we learn how the process really works, describe its directory and write its rules. It joins the same model and works with every other module.
Ask for your own moduleFor governmentOne operating model for the whole agency
- Citizen requests and follow-up
- Budget, procurement, inter-agency work
- Deployed in your environment
For companiesRun the group the way the owner intended
- Deals, procurement, invoices, contracts
- One model for every desk
- A pilot on a few processes first
Your data stays in your country
1 / 4
Your data goes in
Documents, records and conversations go into a vault on your own servers, in your country. Nothing is copied to someone else’s cloud.
2 / 4
A local model works inside
Open models run inside the vault. Routine reading, matching and drafting never leave it, and keep going even with the internet off.
3 / 4
Strong models, masked data
For hard tasks we use the strongest outside models. At the border names, contacts and amounts are replaced, and every call passes through our protection.
4 / 4
The answer comes back whole
The model answers about “Person A”. Inside the vault the real name and amount are put back. The outside service never saw them and keeps nothing.
Virtual reality
Step inside your organisation
Put on a headset and stand in the model: maps on the table, agents at their desks, decisions in the room.
In the room
In the headset
Strategy room: the leadership stands at one table with every map, hypothesis and decision of the company.
Strategy room
Walk through the strategy map with your leadership and decide inside it.
Facilitated sessions
Strategy sessions with an agent as facilitator. Decisions recorded as they are made.
Agent office
See every agent at its desk, what it works on and whom it waits for.
Meeting room
Planning meetings and stand-ups around one table: the agenda, open cases and deadlines in front of everyone, decisions recorded as they are made.
From desert to garden
One seed
We map how work runs today and choose the first process.
First growth
A few processes and their desks run with real people.
It spreads
Your team runs the desks. New units join the same model.
A garden
New processes, desks and agents on subscription.
The people who build it
Questions people ask
What is SORP AI-Native Platform?
A management layer on top of a company’s existing data. It turns documents, systems and conversations into one working model: processes, roles and desks where AI agents prepare the work and people make the decisions.
Do we have to replace our ERP or CRM?
No. The platform connects to the systems you already use, such as CRM, accounting, email and document storage, through standard connectors, and works on top of them.
What do AI agents do, and what stays with people?
Agents read, check, draft and follow up, each with a named owner and set rights. Money, law, signatures and anything sent outside the company are always approved by a person. An agent earns more rights only through a record of approved work.
How does implementation start?
With an assessment of how work really runs today and the choice of one first process. Then a pilot on a few processes with your own people, a handover to your team, and growth on subscription, one process at a time.
Where is our data kept?
Wherever you choose, including your own servers. Local models do the routine work inside your environment. An outside model is used only if you allow it, and then names, contacts and amounts are masked first. The platform keeps working without an internet connection.
Start with one process
Eight things that happen in most organisations. Swipe right if it happens in yours. Your map shows where agents should start.
1 / 8
Drag the card, or use the arrow keys