Artificial intelligence (AI) agents are poised to become a permanent fixture in the business landscape. According to the International Data Corporation, around 70% of organisations in the Asia Pacific region expect agentic AI to disrupt existing business models within the next 18 months. With the capability to act autonomously, these systems are driving a rethink of customer engagement, operational efficiency and the very definition of productivity.
FairPrice Group is already putting this into action. The Singapore-based retailer has developed Grocer Genie, an all-in-one platform for AI-assisted store operations. Powered by Google Cloud’s Vertex AI, BigQuery and Gemini models, the system is accessible via both mobile and web apps. A store manager can simply snap a photo of a spill, and Grocer Genie will automatically assign a cleanup task. It can also provide real-time insights in natural language or visual format when asked about sales or inventory.
To further enhance efficiency, FairPrice Group is set to launch a video analytics solution that integrates Google Cloud’s Vision AI and Gemini with its CCTV systems and Grocer Genie. This will allow AI agents to monitor in-store conditions, detect issues such as low-stock shelves or long checkout queues and alert staff to take immediate action.
These efforts fall under the company’s Store of Tomorrow programme, which aims to elevate the shopping experience through technology. “The last few years of global disruption have shown that the only certainty in retail is how quickly consumer needs, tastes and preferences evolve. Through our Store of Tomorrow programme with Google Cloud and its partner ecosystem, we’re applying interoperable, best-in-class generative AI and data analytics capabilities to reimagine shopper engagement across physical and digital formats and make things easier on the experience and wallet for our customers,” says Vipul Chawla, group CEO of FairPrice Group.
Beyond streamlining internal operations, AI agents are making significant inroads in customer-facing roles. Insurance tech firm bolttech, for example, has developed a generative AI platform built on Amazon Bedrock, supported by Amazon Connect and Amazon Lex, to enhance its omnichannel call centre operations.
This enables bolttech to deploy AI agents capable of handling customer inquiries autonomously — even complex ones — while understanding individual contexts and responding in native languages with personalised policy information. By automating routine tasks such as claims processing, human agents are freed to focus on higher-value interactions without compromising customer satisfaction.
See also: Apple, Google warned by Germany that DeepSeek is illegal
Cloud’s role
While the use cases for AI agents differ between FairPrice Group and bolttech, both early adopters underscore a common reality: cloud computing is foundational to deploying and scaling AI agents. “Cloud and AI agents go hand-in-hand. If you look at frontier AI models, they are trained on cloud resources most of the time, and their evolutions have progressed the capabilities of agentic infrastructures,” Mai-Lan Tomsen Bukovec, vice president of Technology (Data and Analytics) at Amazon Web Services (AWS), tells The Edge Singapore.
An agentic infrastructure is a system in which multiple AI agents with specialised capabilities work together, often autonomously, to accomplish complex goals. Thailand’s Bank of Ayudhya PCL (Krungsri) built such an infrastructure on Amazon Bedrock to accelerate the migration of its data platform from on-premises systems to the cloud.
See also: CoreWeave CEO joins world’s richest after AI firm’s 300% surge
AI agents running on AWS services such as Amazon SageMaker, AWS Lambda and Amazon EventBridge were used to automatically convert legacy code into cloud-compatible formats, cutting migration time by more than 50% compared to manual efforts. Since the AI agents are reusable, Krungsri could maintain consistent and efficient code transformation across multiple projects.
Migrating to the cloud also gave Krungsri the agility to meet new Thai regulations banning mule accounts used in financial crimes like money laundering and fraud. With Amazon SageMaker, the bank transitioned from a reactive approach to a proactive system that automatically detects fraudulent accounts with higher accuracy and fewer false positives.
“Krungsri’s case study shows how AI agents can help organisations get to the cloud, and once they are on the cloud, they gain the ability to use or develop solutions to respond to business and regulatory changes quickly,” says Tomsen Bukovec.
Being on the cloud, she adds, gives organisations access to cutting-edge technologies, specialised AI capabilities and advanced development tools — without the need for significant upfront investment. "AWS is constantly adding new capabilities to our cloud infrastructures and AI models. And more and more services are putting agentic systems or AI agents under the hood. So, in the future, every company is going to be an AI business. While they may not write (or build) the AI infrastructure themselves, they can leverage all our AI solutions or use AWS infrastructure and services as the building blocks of their AI initiatives.”
Philip Weiner, bolttech’s CEO for Asia, agrees. He says: “AWS’s cloud computing and generative AI services, including Amazon Bedrock, provide the foundation to access diverse model choices, deliver superior price-performance ratios, and robust trust and safety enterprise features that align perfectly with our needs. The ability to select from various AI models, such as Amazon Nova, helps support rapid innovation to improve customer experiences in our industry, including how we can offer real-time policy explanations, instant claims processing, and near-human AI interactions by leveraging advances in agentic AI.”
Controlling costs
As businesses expand their use of AI agents across the enterprise, the rising cost of computing power (particularly for inference) has become a growing concern. Inference, the process by which AI models generate real-time responses or predictions, accounts for a significant portion of AI infrastructure costs.
To stay ahead of the latest tech trends, click here for DigitalEdge Section
AWS is tackling the issue by making inference more cost-efficient at scale. According to the company, some of its Amazon Nova foundation models are at least 75% less expensive than top-performing alternatives, while also being the fastest in their intelligence class within Amazon Bedrock. Businesses can test both Amazon’s Nova models and third-party foundation models on Bedrock to determine the best fit for their applications.
Amazon Bedrock also supports model distillation, which is the process of transferring specific knowledge from a more powerful “teacher model” to a smaller, more efficient “student model” that is highly accurate, but also faster and cheaper to run. “With Amazon Bedrock, distilled/student models have less than 2% accuracy loss in use cases like retrieval augmented generation (where the model responds by only referencing a company’s internal or specific data). This means the accuracy of the responses is balanced against the cost of inference,” adds Tomsen Bukovec.
She also pointed to Amazon Bedrock Intelligent Prompt Routing as another tool to optimise both performance and cost. The feature evaluates which model will yield the best quality response for each request and routes the prompt accordingly.
“Organisations will be using different AI models in their agentic infrastructure, as some models are better at a task than others. With Amazon Bedrock Intelligent Prompt Routing, a prompt can be automatically routed to the best model for that particular use case, which gives organisations control over how they want to balance cost and capability/accuracy of their AI initiatives.”
Additionally, AI model providers like Anthropic are now integrating control mechanisms directly into their AI models. Organisations can, for instance, set budgets for “deep reasoning” (where models work through complex problems step-by-step) or opt for “hybrid reasoning” to strike a balance between speed of responses and analytical depth. This level of granularity gives businesses more direct control over performance and cost.
Tomsen Bukovec expects AI systems to serve users across all levels of expertise. “Because of the agentic infrastructure, AI agents will be able to interpret prompts of varying complexity and dynamically construct the right depth of understanding to deliver the best response,” she says. Such adaptability is only truly possible in the cloud, where scalable computing, storage and orchestration enable real-time responsiveness at scale.
