Skip to content

AI agent development

AI agents that finish the job, not just the chat.

ARO Solutions is a software and AI studio in Amman, Jordan that designs and builds production AI agents: software that reads a request, uses your tools and data, and completes a business process from start to finish, with a person approving the steps that matter. It builds with LangGraph, the Claude API and other LLM APIs, and the Model Context Protocol (MCP), in English and Arabic. A first agent is scoped as a fixed-price pilot of about four weeks.

Last updated · ARO Solutions, Amman

What an agent can run for you

An agent earns its place where the work repeats but the input never looks the same twice. Typical first builds:

  • Support and ticket triage

    Reads incoming tickets, emails or chat messages, classifies them, drafts the reply from your knowledge base and routes the rest to the right person.

  • Document checks

    Onboarding and KYC packs, contracts and invoices: extracts the fields, checks them against your rules and flags the gaps for a person to decide.

  • Quote and proposal drafting

    Turns an enquiry plus your price list and past proposals into a draft quote in your format, ready for review.

  • Sales and CRM follow-up

    Keeps records current, prepares the day's call list, drafts follow-ups and writes the weekly numbers.

  • Internal assistants

    Answers staff questions from policies, contracts and records, and shows the source passage for every answer.

  • Scheduled reporting

    Pulls figures from your systems on a schedule, checks them and writes the summary your managers actually read.

Automation, chatbot or agent: which do you need?

They solve different problems. We will tell you when the simpler one is enough.

CriterionRule-based automationChatbot over your documentsAI agent
Best forThe same steps every time, on structured dataAnswering questions from your documentsMulti-step work where the input varies and judgement is needed
Reads free text, PDFs and messagesNoYesYes
Acts in your systemsYes, fixed steps onlyNo, it answersYes, it chooses the steps inside set permissions
Human approvalRarely neededNot needed: it shows its sourcesOn any step that sends, pays or deletes
The right choice whenThe process never changesPeople keep asking the same questionsStaff spend hours reading, deciding and copying between systems

Swipe the table sideways to see every column.

How we build agents that hold up in production

  • One process first, measured against success criteria agreed before we start: for example accuracy on 100 real test cases, or minutes saved per ticket.

  • A person approves every step that sends a message, moves money, deletes data or commits the business.

  • Tools are exposed through MCP servers and typed APIs, so the agent can only do what it was given permission to do.

  • An evaluation set built from your real cases runs on every change, so quality is a number and not an impression.

  • Every run is logged: what the agent read, which tool it called, what it answered and why.

  • Claude by default, other LLM APIs where they fit better, and in-region or self-hosted models when data rules require it.

  • Arabic and English from the first version, including dialect input and right-to-left interfaces.

How a pilot runs

  1. A 30-minute call: the process, the systems it touches and what success would look like.

  2. Within 48 hours you get the scope, a fixed price and a timeline in writing.

  3. About four weeks of building in your environment, with a demo every week.

  4. Handover: the evaluation report, the code in your repositories and the documentation. Then we extend it to the next process, or you take it from there.

Agents we run ourselves

HerShape, the AI fitness app ARO built and operates, includes Coach Sarah: an agent that knows the user's plan and progress log and rewrites a meal or a training session from one chat message, instead of only answering questions. It is live on the App Store and Google Play.

We also run parts of our own studio on agents we built: lead research, CRM updates and the weekly numbers.

What we build with

  • LangGraph
  • Claude API
  • MCP
  • Python
  • RAG
  • Vector databases
  • Postgres
  • Next.js

Questions people ask

What is an AI agent, in plain terms?

An AI agent is software built on a large language model that can read a request, decide which steps to take, call tools such as your CRM, email or database, and return a finished result. A chatbot answers a question; an agent completes a task. ARO builds agents that run one defined business process and ask a person to approve the sensitive steps.

How long does it take to build an AI agent?

A first agent that runs one process end to end is scoped as a fixed-price pilot of about four weeks, including the evaluation report. Scope, price and timeline are confirmed in writing within 48 hours of the first call.

How much does an AI agent cost?

ARO quotes a fixed price for the pilot after a short scoping call, instead of billing open-ended hours. The price depends on how many systems the agent connects to and how the result is evaluated. The running cost afterwards is the model usage, billed by the provider to your own account.

Which frameworks and models do you use?

LangGraph for orchestration, the Claude API and other LLM APIs for reasoning, MCP servers for tool access, Python, and Postgres with a vector database for retrieval. When a client's rules require it, we use in-region or self-hosted models.

Do you build MCP servers?

Yes. We build Model Context Protocol servers that expose your internal systems (databases, CRMs, document stores, internal APIs) as tools an AI assistant can call with scoped permissions. The same integration then works in Claude and in any other MCP-compatible client.

Can the agent work in Arabic?

Yes. Agents read and write Arabic and English, handle questions asked in dialect against documents written in formal Arabic, and work through right-to-left interfaces. Arabic document work is covered in detail on the Arabic AI page.

Where does our data go?

It stays in your environment. The code lives in your repositories, we document which LLM APIs are called and their data-retention settings, and secrets are kept in a secrets manager. In-region or self-hosted models are available when your rules require data to stay in the country.

Can a company outside Jordan hire ARO?

Yes. ARO is based in Amman (UTC+3) and takes projects remotely, in English and Arabic, from the Gulf, the UK, Europe and North America. Amman is on the same time as Riyadh, one hour behind Dubai, two to three hours ahead of London and seven to eight hours ahead of New York.

Got an idea?
Let's build.

Tell us what you want to build. We usually reply within a few hours. Worst case, by the end of the day.

Amman, Jordan · Taking projects worldwide

Free intro call

Not sure what you need yet? Book 30 minutes — we'll figure out what's actually worth building.