Amrak turns one sentence from someone on your team into finished work inside the systems your company already runs. Every request passes through the same five steps: Amrak understands what was asked, checks that the person is allowed to do it, prepares the full change, waits for approval if the action is sensitive, then completes the task and logs it. Behind those steps is a connector we build for each company on MCP, an open standard for linking AI assistants to business systems, so the assistant works with your own sections, your own rules and each person's permissions. Nothing moves to a new platform and no data is migrated. In this article we open the box and follow real requests from the first word to the final log entry.
Why did we build Amrak around the systems companies already have?
Because the gap in most companies is not access to AI. It is the distance between the chat window and the place where the work is actually done. McKinsey's survey of senior executives and board directors in the Gulf found that 84 percent of GCC organisations had adopted AI to some extent in 2025, up from 62 percent in 2023. Yet only 31 percent said AI was being scaled or fully deployed across the organisation, and only 11 percent qualified as "value realizers" that can attribute at least 5 percent of earnings to AI.
The global picture is similar. In McKinsey's State of AI in 2026, nearly nine in ten respondents report regular use of AI in at least one business function, but only 37 percent attribute any EBIT impact to it.
We see the same pattern in our own work and in the companies we talk to. Someone asks an AI tool for a draft, copies it, opens the CMS or the CRM and does the rest by hand. The AI saves a few minutes of writing and none of the clicking. So we built Amrak the other way round: the request starts in plain Arabic or English and ends inside your system, done, checked and recorded. We introduced the service in our Amrak launch announcement. This article goes under the hood.
What happens between pressing send and the task being done?
Each request goes through five steps, in the same order every time. The table shows what Amrak does at each step and what it means for the person who asked.
| Step | What Amrak does | What it means for you |
|---|---|---|
| 1. Understands | Works out the task, the system it belongs to and the details it needs | You write the way you talk, in Arabic or English, with no forms to fill |
| 2. Checks | Confirms that your permissions allow this action | You can only reach what your role allows |
| 3. Prepares | Builds the full change before anything is saved | The work is complete, not half done |
| 4. Asks you | Stops and waits if the action is sensitive | Messages to clients, deletions and price changes never happen without approval |
| 5. Does it | Completes the task and records it | The log shows who asked, when and what changed |
The steps sound simple, and that is the point. The intelligence is in the first and third steps. The safety is in the second, fourth and fifth.
How does one sentence become a scheduled article?
Take a marketing manager who writes: "Schedule the article about our new delivery service for next Sunday at ten in the morning, use the warehouse photo from the image library, and fill in the search title and description."
- Understands. Amrak reads four things in that sentence: which article (the draft about the delivery service), when (next Sunday, ten in the morning), which image (the warehouse photo already in the library) and which extra fields (the search title and description).
- Checks. The manager's role allows publishing on the blog, so the request continues. A colleague whose role covers drafting only could prepare the article but could not schedule it.
- Prepares. Amrak assembles the full entry: the article text, its category, the chosen image, a search title and description that fit, and the publication date and time.
- Asks, if needed. Scheduling an article is not one of the sensitive actions such as messaging clients, deleting content or changing prices, so it goes ahead. A request to delete an old article would stop here and wait for approval.
- Does it and logs it. The article is scheduled in the CMS, and the log records who asked, when and what changed. The manager reviews the result.
On our Amrak page we count 14 manual steps to publish one article by hand: logging in, finding the right section, then field after field for the title, text, image, category, search details and date. Here the person writes one sentence and checks the result. The work did not disappear. The clicking did.
What happens when a request touches money or clients?
The same five steps behave differently when the stakes rise. Take an operations manager who writes: "Which items are below their minimum stock level? Draft a purchase order for them."
- Understands: two tasks, a question and a draft, both in the ERP.
- Checks: her role lets her read stock levels and create draft purchase orders, but not change prices. This is the kind of split the command map defines for each role, just as a person may be allowed to edit products but not prices.
- Prepares: a list of the items below their limit with current quantities, and a draft order built from that list.
- Asks: saving a draft commits nothing. If she had also asked to update an item's price, that part would stop and wait for approval, because price changes are always treated as sensitive.
- Does it: the draft is saved in the ERP and the action is logged. Issuing the order stays with whoever holds that permission.
A sales request works the same way. "Draft follow-ups for clients we have not contacted in a month" produces ready drafts, but messages to clients always wait for approval, so nothing reaches a client until a person says yes. The rule underneath every example is simple: Amrak can only reach what the person asking can reach, and anything sensitive waits for a human.
Which systems can Amrak work in?
Amrak works inside the systems you already have. The connector we build covers your own sections and rules, and what Amrak can do in each system depends on what that system allows, which we confirm with you in the discovery session.
- Website and content publishing: articles, pages, portfolio, product store and image library. Change a phone number on every page, or add a product, a project or an FAQ.
- Dashboards and analytics: explain a drop in traffic, show the top-performing pages or summarise yesterday's work.
- CRM: look up clients and draft follow-ups that wait for approval.
- ERP and operations: flag low stock and draft purchase orders.
- Calendar and scheduling: set publication dates or reschedule many posts at once.
- Client portal and design tools, for tasks such as designing a booklet page.
There is no data migration and no system replacement. If you are still deciding which system should hold which data, our CRM vs ERP guide is a good place to start.
Why is Amrak built on MCP?
MCP, the Model Context Protocol, is an open standard for connecting AI assistants to the systems where company data lives. Anthropic released it as open source on 25 November 2024. We explained how it works in what MCP is and how it connects AI to your systems, so here we only cover what it means for an Amrak client.
- The connector is built around you. We build it for your sections, your words and your rules, with the commands and role permissions agreed in your command map.
- The AI provider is your choice. Because the connection follows an open standard, you choose the provider that fits your company's policy.
- Your system stays the source of truth. Amrak works on top of what you already run. Your data stays where it is.
We wrote about why that independence matters in how to avoid AI vendor lock-in.
What stays under your company's control?
Everything that matters. These are the rules every Amrak connector follows:
- Approval for sensitive actions. Messages to clients, deleting anything and changing prices always wait for approval.
- Permissions per person. Each person reaches only what their role allows.
- A log of every action. Who asked, when and what changed.
- Your data stays in your system. Amrak reads only what each request needs.
- Your choice of AI provider. You decide who processes your requests.
- An off switch. You can switch Amrak off at any time.
An upcoming article in this series is entirely about these controls: how to set permissions by role, which actions should wait for a person and what a useful log looks like.
How does an Amrak project run, and how is it priced?
We work with every company in six stages:
- Free discovery session: we learn your business, your systems and the tasks that take your team's time, with no commitment.
- Command map: together we choose the most valuable commands and the permissions for each role.
- Build: we build the connector to fit your system.
- Testing with your team: your team uses it on real tasks and we refine it from their feedback.
- Handover and training: your team learns to write good commands and gets a ready command guide.
- Support and growth: regular follow-up, and new commands as your work grows.
Timelines for each stage are agreed in the quotation. The model has two parts. A one-time build fee covers studying your system, building the connector, testing and training. A monthly subscription covers technical support, regular follow-up and AI image generation when needed. Extra commands or systems are quoted before any work starts, and existing HBS clients get special terms. To judge the result in your own numbers, use the metrics in how to measure AI ROI.
What will the rest of this series cover?
This article is the overview. In the coming articles we go deeper: ready-to-use commands for each department and how to write a good one, how we run our own website, articles, marketing calendar and client portal on Amrak, and the safety controls in detail. Then we look at the everyday problems that lead companies to Amrak, with the numbers behind each one and steps you can take with or without us: the hidden time cost of clicking through dashboards, why teams avoid the CRM, adding AI without replacing your ERP, updating your website without waiting for a developer, publishing content at scale with a small team, and helping new hires become productive in their first week.
If you want to see what one sentence could do inside your own systems, book a free discovery session. Tell us which task takes most of your team's time, and we will map the commands that would handle it, with the permissions and approvals your company needs.
