The short answer
We are building an agentic platform for organizations in Jordan and the region that need AI to work on their own data, in their own language and under their own rules. Four principles shape it: Arabic-first, data kept in the country, governance built in and a person approves every action that matters.
Why a platform, not one agent at a time
In our previous article we noted Gartner's forecast that over 40% of agentic AI projects will be cancelled by the end of 2027, often because of unclear value and weak risk controls.
Building agents one by one repeats the same hard work each time: connecting data, setting permissions, logging and approval. A platform does this once, so every new agent starts governed.
1. Arabic-first
Most AI tools are built in English and translated later. We start from Arabic:
- Questions in Arabic or English, including mixed sentences, the way people speak at work.
- Local business terms mapped to one agreed definition, so a figure means the same thing in every answer.
- Answers written in Arabic, not translated from English.
- Right-to-left screens designed that way from the start.
2. Data stays in the country
Jordan's Personal Data Protection Law (No. 24 of 2023) took effect in March 2024 and limits transfers of personal data abroad unless adequate protection is in place. For banks, telecom operators and public bodies, where data is stored and processed is a first question, not a detail.
Our design follows the sensitivity of the data:
| Data | Where it is stored | Where the AI runs |
|---|---|---|
| Public (open statistics, market data) | Cloud | Cloud AI services |
| Internal (operations figures, no personal data) | Separate for each customer, in the region | Cloud services that keep no data, or models inside the country |
| Regulated (customer personal data, banking, telecom) | In the country only | Models running in the country only |
3. Governance built in
Governance is part of the first release, not an add-on. Every request passes through the same controls:
- Agreed definitions: agents read data only through approved business definitions, never raw tables.
- Access follows the person: an agent sees only what the person asking is allowed to see.
- Personal data masked before any text leaves the customer's environment.
- Full audit trail: every question, data request, step and answer is logged and can be replayed.
- Cost limits for each customer, with alerts.
4. A person approves
Agents prepare; people decide. Any step that writes to a system, sends a message outside the organization or touches regulated data waits for a named person. This matches the direction of regulation: the EU AI Act (Article 14) requires high-risk AI systems to let people oversee them and stop or override them.
Examples of agent work, with a person approving:
- Scheduled report: the agent prepares a daily summary in Arabic — the owner reviews it before it is sent.
- Alert and act: the agent finds an unusual figure, explains it and drafts a ticket — a person approves it.
- Approval chain: the agent drafts the step, a person approves, then the agent carries it out.
Built on open standards
Agents connect to data and tools through the Model Context Protocol (MCP), and to agents in other systems through the Agent2Agent protocol (A2A). Both are open standards hosted by the Linux Foundation. A2A joined in June 2025 and had more than 150 supporting organizations by April 2026.
For a customer, this means the platform works alongside the systems already in use instead of replacing them.
What comes first
The first release does one job well: ask a business question in Arabic or English and get a governed answer, with an explanation, a saved dashboard and scheduled reports. A workspace where teams build their own agents from templates follows later.
Our view
Agentic AI will reach the region either as tools built elsewhere and adapted, or as platforms built here for local language, law and accountability. We are building the second. Our principle stays the same: AI provides the analysis. The person owns the decision.
Interested in an early pilot? Book a 30-minute call or see how our pilots work.
Sources
- Gartner, Gartner Predicts Over 40% of Agentic AI Projects Will Be Canceled by End of 2027 (25 June 2025)
- Clyde & Co, Jordan issues first personal data protection law (October 2023)
- European Commission, AI Act Service Desk, Article 14: Human oversight
- Linux Foundation, Linux Foundation Announces the Formation of the Agentic AI Foundation; Model Context Protocol a founding project (9 December 2025)
- Linux Foundation, Linux Foundation Launches the Agent2Agent Protocol Project (23 June 2025)
- Linux Foundation, A2A Protocol Surpasses 150 Organizations (April 2026)