E-commerce in Saudi Arabia is heading to a stage in which competition exceeds the speed of executing the payment process or adding new means of payment, to the point of making the payment itself less visible to the consumer, coinciding with the rise of artificial intelligence agents capable of searching, comparing, and taking purchasing steps on behalf of the user.
Remo Giovanni Abbondandolo, General Manager of the Middle East and North Africa at Checkout.com, believes that the Saudi market is likely to adopt this transformation quickly, based on the Kingdom’s record in adopting new payment technologies, but at the same time he stresses that trade through artificial intelligence agents still raises basic questions about authenticating the transaction, responsibility for errors, and how to manage refunds and objections to operations.
Abundandolo said, in an interview with Asharq Al-Awsat on the sidelines of the LEAP 2026 conference in Riyadh, that the company is particularly following developments in “Agent Commerce” at a time when a large part of the technical discussion at the conference is focused on artificial intelligence and its new applications.
98% want less visible payment
According to Checkout.com data, 98 percent of Saudi consumers see value in payment experiences that are not visible or more integrated into the purchasing journey. Abundandolo says that the Saudi consumer is “very mobile friendly” and places great importance on the user experience, while the ideal payment process for him becomes smoother and less frictionless. But getting to this point is not about rest alone. According to the numbers he referred to, 31 percent of Saudi consumers said that they had been exposed to potential fraudulent attempts or transactions, while 27 percent reported that a payment was rejected even though the card was working, a situation known in the payments sector as false rejection.
He stated that the challenge for merchants and payment providers is to achieve a balance between raising transaction acceptance rates and reducing friction, without this coming at the expense of protection and trust. He added: “There is still a lot of work to improve the performance of payments, but we see that it is improving year after year,” pointing to the role of direct integration with payment networks, alternative payment methods and digital wallets in this development.
AI agents enter the buying journey
The biggest shift may come from artificial intelligence itself. According to company data, more than half of Saudi consumers are willing to allow an artificial intelligence agent to shop on their behalf. Abbondandolo expects this model to turn into an additional channel for e-commerce, alongside websites, apps and social commerce, rather than completely replacing existing channels. He stated: “We will see more consumers using artificial intelligence not only to compare prices, read reviews and get advice, but also to make purchases.”
He bases his prediction on the Saudi market’s rapid adoption of previous payment technologies. He pointed out that the introduction of “Mada” into e-commerce witnessed rapid growth, and in one of the first stages it reached about 30 percent of transactions, according to what he mentioned, while Saudi Arabia was one of the markets that recorded rapid growth in the use of “Apple Pay.” He believes that agent trade may follow a similar path, but the main problem is that its operational and legal structure is not yet complete. He added that the sector still needs to address issues related to “security, how to deal with refunds and objections, and how to direct the agent,” which he said payment companies, card networks and technical bodies need to solve jointly.
Who approves the deal?
The problem becomes more complicated when a human is not the one directly clicking the purchase button, as current digital payments aim to verify a person's identity.
In the case of the artificial intelligence agent, the question becomes twofold: Has the identity of the account holder been confirmed? Did he specifically approve of this operation carried out by the agent?
“Authentication is one of the aspects that still represents a challenge in agent trading,” Abbondandolo answers, explaining that the merchant needs to ensure that the person who gave instructions to the agent is fully authenticated, and also that the specific transaction falls within the limits of the authorization he granted. He gives a simple example: If an agent buys shoes that are not what they wanted or books the wrong flight, who bears responsibility?
In agent trade, the issue of authenticating the agent's decision itself still requires the development of new rules and mechanisms.
Artificial intelligence confronts fraud
On the other hand, artificial intelligence is already entering the current payments architecture, especially in detecting fraud and improving the acceptance of transactions. Abbondandolo pointed out that data-based tools and machine learning can help more accurately distinguish between a fraudulent transaction and a valid transaction that may be rejected by traditional rules of protection.
He said that Checkout.com uses a system based on artificial intelligence to analyze the chances of the operation being successful and to retry some operations that were subject to infinite rejection. In his opinion, through these mechanisms, the company recovers the equivalent of a billion dollars every 30 days from its customers’ transactions that failed at first and then their path was improved and completed.
Digital wallets are reshaping behavior
These shifts come in a market where digital wallets are becoming an increasing part of daily financial behavior. Abbondandolo attributes its spread to three main factors: the ease of paying using the phone, the authentication provided by wallets, and the use of encryption techniques that prevent the original card number from being passed directly in every transaction.
He said that transactions conducted through coded cards provide the company with better performance than some traditional payment operations. But this does not mean the disappearance of cards or other local solutions, but rather the addition of another means through which the consumer chooses the method that suits him.
Digital identity is at the heart of the next stage
As the card itself becomes less visible within the payment journey, it becomes increasingly important to know the identity of the person behind the wallet or agent. Abundandolo points out that the digital identity will play a greater role because it helps verify that the user is actually the owner of the wallet or account from which the payment is made.
Traders need to prepare for a new channel
For Saudi retailers, preparing for proxy commerce is not just about updating the checkout page. The agent can compare prices, reviews, and offers faster than the user, which means that the appearance of the product, its information, and its purchase terms will be read and evaluated by automated systems.
2030: When electronic payment becomes like a store
When asked what a successful Saudi payments system will look like by 2030, Abundandolo sets a simple goal: for the electronic payment experience to become as close as possible to the in-store payment experience.
On Thursday, OpenAI unveiled a new model of artificial intelligence, which it described as the best so far, but warned that it sometimes tries to evade human oversight, at a time when the company faces increasing scrutiny after its agents penetrated the systems of other companies.
OpenAI is suffering from the repercussions of the incident that occurred in July, when its agents managed to evade a secure test and hacked into the systems of the open source Hacking Face platform, in an attempt to cover their tracks. This incident has raised safety concerns, as similar incidents have occurred at rival Anthropic as developers race to deploy more advanced models.
Concerns center on agent-based artificial intelligence, designed to perform tasks with little or no human intervention. The promise of running agents around the clock is a key factor in investors' confidence in the appeal of artificial intelligence as a transformative technology.
OpenAI named its latest model GPT-6 Astra, which follows the release of GPT 5.6 Soul in July, and said in a blog post that it is faster and capable of performing more tasks than any previous version. Astra's skills include preparing tax returns, developing games, preparing architectural designs, coordinating legal memos, and searching for residential apartments.
“Astra represents a new horizon in speed, accuracy and security in computer use,” the company said. Greg Brockman, president of OpenAI, said in a briefing earlier in the day that Astra represents “a real shift in the type of work people can delegate to AI and how it can empower them.”
For example, the company said that Astra reduced the time required to search for a cat sitter from 30 minutes, when done by a human, to 5 minutes and 27 seconds. As for searching for a job, it took only 2 minutes and 51 seconds, compared to five hours without the Astra.
However, OpenAI also reported that Astra is more likely to deliberately hide or camouflage its detailed or step-by-step methods for solving problems, known as “heuristics,” making it difficult for humans to evaluate its techniques later. As for more complex problems, OpenAI said Astra cannot yet conceal its methods consistently, although it is getting better at concealing its effects.
In a briefing Thursday morning, OpenAI's chief expert noted that monitoring is becoming more difficult, along with harmonization, the principle that AI should reflect human values.
“As models become more capable, it becomes more difficult to understand exactly what they can do,” Jakob Paczocki said. “This is no guarantee that our methods will be adequate as intelligence continues to increase, because advances in intelligence do not guarantee advances in fitness.”
Monitoring AI agents is a key component of OpenAI's assurances to regulators, legislators and the public to help avoid another security incident. Moreover, the company this week informed two Democratic US House members in a letter that it is developing “automated shutdown capabilities” for its models.
Astra can also help companies discover vulnerabilities in their systems more quickly, but it also makes it “easier to exploit those vulnerabilities,” OpenAI said. The company added that for these reasons, it may have to conduct additional security checks that “can slow, temporarily stop, or sometimes terminate legitimate business, including defensive cybersecurity.”
OpenAI said last month it had temporarily halted some model development work, partly to ensure it could be monitored. The company's chief expert said that concern that models would develop capabilities to disable monitors or evade them entirely was "very understandable," adding that OpenAI was trying to address these problems.
OpenAI is striving to catch up with rival Anthropic among corporate clients, as rival Anthropic gains market share ahead of its widely anticipated IPO later this year.
Astra targets a wide range of enterprise customers, who OpenAI said it hopes will be attracted by its speed and versatility. It is available today to a limited group of customers, and will be rolled out more widely in the coming days.
