AI agent development and agentic AI for business

An AI agent is a language model given instructions, access to your data and a set of tools, so it can work through a task rather than answer one question. Anthropic draws a useful line: workflows follow predefined paths, while agents decide their own steps and tool use. Agentic systems suit open-ended work, but they often trade speed and cost for better task performance, and their errors can compound, so the simplest design that does the job is usually the right one.

For organisations in Saudi Arabia, the UAE, Qatar, Jordan, Kuwait and Oman, good first candidates are the processes where people move information by hand between email, documents and the ERP: invoices, purchase order follow-ups, reconciliations, service cases and approvals. These processes often run in Arabic and English and need an audit trail. So we build agents into the process, not beside it: AI and agents sit between trusted data and the roles and controls that approve what happens next.

Microsoft now ships agents inside Dynamics 365, so agentic ERP can start in the system you already run. Many of these agents carry a preview or production-ready preview label, and availability varies by region and language. We check each one against your version, licences and data settings and configure its rules and limits. Then we test it in a sandbox before it reaches production, the path Microsoft describes for its Sales agents.

Agents Microsoft documents today include:

• Finance: the Account Reconciliation Agent, which evaluates reconciliation exceptions and recommends an action for each
• Supply Chain Management: the Procurement Agent, whose supplier communications features draft follow-ups to vendors and match their replies to purchase orders
• Project Operations: the Time Entry, Expense and Approvals agents
• Business Central: the Payables Agent, Sales Order Agent and Expense Agent
• Sales: the Sales Qualification, Sales Opportunity, Sales Close and Sales Research agents
• Customer Service: the Case Management, Customer Intent and Customer Knowledge Management agents

• Dynamics 365 Finance & Operations
• Dynamics 365 Business Central
• Dynamics 365 CRM

When the built-in agents do not fit your process, we build your own. For finance and operations apps, Microsoft’s Dynamics 365 ERP MCP server lets an agent read and update data, work through application pages and call business logic, but only within the security role it is given; calls outside that role are rejected. Microsoft’s own guidance for agents built on it in Copilot Studio recommends a Claude model for orchestration. ConsultTik builds AI solutions on Claude as well as on Azure OpenAI and Copilot Studio, and chooses per use case.

We choose the platform by the work, the users and your governance needs:

• Copilot Studio for low-code agents in Microsoft Teams, Microsoft Copilot, your website or your mobile apps, with connectors, agent flows and human review steps
• Microsoft Foundry Agent Service for agents that need code-level control, with a dedicated Microsoft Entra identity per agent, private networking, content filters, tracing and evaluations
• Anthropic’s Claude models, available in Microsoft Foundry and selectable in Copilot Studio, for complex reasoning and long documents
• Model Context Protocol (MCP), an open standard for connecting agents to systems, including the Dynamics 365 ERP MCP server

• Copilot Studio agents
• Claude for enterprise

Good first agents remove repetitive handling without taking decisions away from the people accountable for them. If accounts payable is your starting point, our AP efficiency and agentic AI assessment estimates where the manual effort sits, finds the main bottleneck and suggests a first agent pilot. It takes a few minutes, is in English and needs no ERP access.

Typical starting points we discuss with finance, operations and service teams:

• Accounts payable: invoice intake and validation, purchase order matching with exception triage, approval follow-up and supplier enquiries
• Procurement: follow-ups on unconfirmed or late purchase orders, with changes proposed for a buyer to accept
• Finance close: reconciliation exceptions explained and routed to the right owner
• Customer service: cases created and updated from email, and knowledge articles drafted from resolved cases
• Employees: answers on policies and procedures, in Arabic and English

• AP efficiency and agentic AI assessment
• Invoice capture and AP automation

Our principle is AI with control: AI acts within roles and approvals; people stay accountable. Microsoft’s agents follow the same pattern. The Procurement Agent in Supply Chain Management does not send supplier emails automatically unless an administrator turns that on, and the Account Reconciliation Agent logs every action that users, automation or agents take on an exception. Our own product shows it too: in DONOR 360, AI drafts and people approve, and the AI voice agent says it is virtual and hands over to a person on request.

In practice, every agent we build has:

• An identity of its own or the signed-in user’s, with the narrowest security role that lets it do its job
• A clear line between what it may do alone and what it must propose for approval
• A record of what it read, proposed and changed, so finance and audit teams can trace each action
• Data policies and admin settings that limit the connectors, channels and models it may use
• Testing in a sandbox with real cases before go-live, and monitoring afterwards

• Our approach: AI with control
• DONOR 360

Language comes first. Microsoft lists Arabic for Copilot Studio’s generative answers, orchestration and voice, and Claude works in Arabic as well; Anthropic’s multilingual benchmarks include it. Some components are English-first: the Dynamics 365 ERP MCP server returns its metadata and guidance in US English, even when labels and values appear in the user’s language. We write instructions and test cases in both languages and test with Arabic-speaking users.

Data residency comes next. Microsoft states that Dynamics 365 Sales agents may process and store data outside the user’s primary region, and that once cross-region data movement is allowed, prompts from environments hosted in the UAE may be processed by the Azure OpenAI service in the United States. In Microsoft Foundry, Claude models run either on Azure or on Anthropic’s own infrastructure, depending on the model and the hosting option you choose. All are offered through Global Standard deployments, and some Azure-hosted models also through US Data Zone deployments. We map where each agent’s data goes and agree the settings with your IT and compliance teams.

Then the law: Saudi Arabia’s Personal Data Protection Law, the UAE’s Federal Decree-Law No. 45 of 2021, Jordan’s Personal Data Protection Law No. 24 of 2023, which sets conditions for transferring personal data outside Jordan, and any personal data rules that apply to you in Qatar, Kuwait or Oman. We design agents with them in mind and leave legal interpretation to your advisers.

Agents are only as useful as the systems they act on, and we implement those systems: Dynamics 365 for ERP and CRM, Power Platform, Microsoft Fabric and Azure. ConsultTik is a Microsoft Solutions Partner with designations in Data & AI (Azure), Business Applications, Digital & App Innovation (Azure), AI Business Solutions and Cloud & AI Platforms, with 15+ years of experience.

We deliver in Arabic and English from offices in Jordan, Qatar and Saudi Arabia and serve organisations in Kuwait, the UAE and Oman. One relationship covers the whole lifecycle, from advice and design through implementation and adoption to managed services with 24/7 support, so the team that builds your agents also helps you run them.

• Microsoft Solutions Partner
• Managed services