AI's next frontier isn't the chatbot but the back office of the physical economy. Dubai startup aradus deploys AI agents that automate routine operations for manufacturers and distributors on top of their existing systems, keeping humans on key decisions, part of a broader shift from AI that talks to AI that works.

[For more news, click here] 

Most of what the public sees of artificial intelligence still happens in a chat window. The more consequential shift may be unfolding somewhere far less visible, inside the order desks, warehouses, and finance teams of the companies that make and move physical goods, where software has begun quietly doing the operational work that keeps commerce running.

A manufacturer or distributor runs on a relentless stream of small tasks such as orders arriving by email and WhatsApp. Invoices need matching, shipments need tracking, documents need validating, and suppliers and customers need chasing. Much of that information sits scattered across enterprise systems, and much of the working day goes to finding it, re-entering it, and following up by hand. It is the machinery of the physical economy, and the kind of work a new class of AI is being built to absorb.

aradus, a Dubai-based startup launched in May 2026, is one entrant aiming to pawn off repetitive manual tasks to free up human time. Rather than a chatbot bolted onto a website, it deploys AI agents that carry out routine operational tasks, reading and validating incoming documents, managing purchase orders and inventory, tracking shipments, following up on missing paperwork and keeping records current. Crucially, it does this on top of the systems a company already runs, its ERP, transport or warehouse management software, rather than asking it to rip those systems out. When a task needs human judgment or sign-off, the platform routes it to a team member with the context and a recommended action attached, and handles the rest once a person approves.

The company says the approach is already producing results in live deployments. According to aradus, its agents have automated more than 90 per cent of customer orders and inbound inquiries for clients across the MENA region, and its shipment tracking alone avoids an estimated $100 to $200 in demurrage charges per container by keeping cargo from sitting past its free time. Those figures are the company's own, drawn from its customer workflows rather than independent audit. The direction the results point to is consistent with where enterprise AI is heading.

"Manufacturers and distributors across different sectors deal with thousands of operational tasks to keep the goods moving every day. We built aradus to take the burden off their teams by providing an AI workforce that can execute routine operational tasks across the systems they already rely on daily. By automating and streamlining everyday workflows and only bringing people into the loop when human judgement is required, we're helping businesses operate faster, smarter and more efficiently so that their teams can focus on the work that truly requires their attention," said Karam El Assaad, co-founder and CEO of aradus.

The Frontier Beyond the Chatbot

What aradus is chasing is the next phase of the AI story, the move from tools that generate text to agents that take actions. Gartner expects that by 2028, a third of enterprise software applications will include agentic AI, up from less than 1 percent in 2024, and that 15 percent of day-to-day work decisions will be made autonomously by such systems. Much of the early attention has gone to knowledge work, but the larger and less-digitized target is the physical economy, the procurement, logistics, order management, and finance operations that sit behind every shipped parcel.

That target is attractive for a simple reason, the work is high-volume, rules-heavy, and repetitive, which is what agents handle best, and it has been resistant to the previous generation of automation because so much of it lives in unstructured places, an emailed order, a scanned certificate, a WhatsApp message from a supplier. For the enterprise-software industry, including the incumbents that dominate US corporate back offices, this is fast becoming the contested ground: not whether AI can converse, but whether it can reliably execute the operational tasks a business depends on.

Building on Top, Not Ripping Out

The design choice that matters most in aradus's pitch is that it runs on top of existing systems rather than replacing them. Anyone who has lived through an ERP migration understands that rip-and-replace projects are notoriously expensive, slow and risky, and the prospect of another one is often what stops a mid-sized manufacturer from modernizing at all. An operational layer that leaves the system of record intact and simply acts across it lowers the barrier to adoption considerably.

It also concentrates the difficulty. Sitting on top of a company's real systems means confronting its real data, and operational data in manufacturing and distribution is famously messy, spread across legacy software, inconsistent formats, and years of workarounds. The quality of that data, more than the sophistication of any model, tends to decide whether this kind of AI delivers. It is the same lesson emerging across enterprise AI: the model is rarely the hard part, and the integration and data foundation almost always are.

The Distance Between a Demo and Production

While impressive AI demonstrations are common; durable production systems are not. In a widely cited 2025 study, MIT researchers found that roughly 95 percent of enterprise generative AI pilots produced no measurable impact on profit and loss, with poor integration into real workflows, not weak models, identified as the culprit. Against that backdrop, a claim to be running in production and automating the bulk of a client's order flow is exactly the differentiator that would matter, if it holds at scale across many customers rather than a handful of favorable early ones.

The broader agentic wave carries its own warning. Gartner has predicted that more than 40 percent of agentic AI projects will be canceled by the end of 2027, done in not by the technology but by escalating costs, unclear business value, and inadequate risk controls. aradus's human-in-the-loop design, keeping people on the consequential decisions and letting agents handle the routine, is a sensible hedge against the autonomy risks that sink such projects. Whether it translates into lasting value will depend on the unglamorous work of integration, data hygiene, and earned trust, not on any single statistic.

Why the Gulf is Fertile Ground

There is a reason a startup like this is emerging from Dubai. The Gulf's economy is unusually physical, built on trade, distribution, logistics and an aggressive manufacturing diversification push, which makes operational AI a natural fit. National AI strategies and heavy enterprise investment are accelerating adoption, and PwC has projected that AI could add $320 billion to the Middle East economy by 2030, equivalent to 11 percent of regional GDP. A trade hub that moves enormous volumes of goods is precisely where automating the paperwork and the follow-ups can pay off fastest.

The same caution applies there as anywhere where the speed of adoption is not the same as depth of payoff, and a region investing quickly can reach the hard operational questions sooner rather than avoid them. What the aradus launch signals is less about one company than about where the contest in enterprise AI is moving. The question is shifting from whether AI can hold a conversation to whether it can reliably do the work.

Related Articles

Amber Lands €7M to Turn Enterprise Knowledge into Autonomous AI

Cequence Security Launches Agent Personas to Automate AI Agent Governance Across the Enterprise

How Can Gulf Enterprises Measure Their AI Security Readiness? SANS Institute Has a Framework