[DigitalToday reporter Chi-gyu Hwang] "2026 is the year of AI agents. The biggest differentiator in AI agents is data."

Rahul Pathak (라훌 파탁), vice president for data and artificial intelligence at Amazon Web Services (AWS), on Tuesday described this year as the first year of AI agents. He presented four priorities drawn from more than 1,000 conversations with customers and partners at a morning news briefing.

The keywords Pathak cited were goals, data, guardrails and execution speed. He explained these are the core factors that determine whether an enterprise AI project moves from a pilot to deployment.

Pathak stressed that AI projects should focus on important tasks and be designed from the start with deployment in mind. He said companies also need a data-use system for AI and security boundaries that let employees work safely without hindering innovation. He added that results must be delivered in weeks, not months, and that failure to produce outcomes quickly can leave companies behind as markets change fast.

He cited Accenture, 1CloudHub and Capgemini as examples and said these partners built agentic AI platforms based on Amazon Bedrock and Amazon Bedrock AgentCore. He said they recorded productivity improvements of 50 to 85 percent across finance, healthcare, manufacturing and insurance and achieved 4 times return on investment (ROI).

Pathak said AWS has the broadest platform to support AI agents and introduced the AWS agentic AI stack underpinning an AI agent strategy. He said the infrastructure layer includes Trainium and GPUs, and the data layer includes Amazon S3 data lake, Redshift, Glue and Lake Formation. He said Amazon SageMaker supports custom machine learning training, deployment and MLOps. He said Amazon Bedrock provides foundation models, fine-tuning, retrieval-augmented generation (RAG) and knowledge bases. He said AgentCore covers orchestration, memory, guardrails and tool use.

Pathak said AWS is investing aggressively in AI and is pursuing deep integration with Anthropic, OpenAI, Nvidia and Cerebras. He said Bedrock supports OpenAI integration and support for new frontier models, and that Anthropic Claude Opus 4.7 is also provided through Bedrock.

Pathak also stressed that workflows based on diverse models and tools, rather than a single model and a single tool, are spreading. "The era of using one agent with one model is over. It is being implemented in multi-tool-based pipelines in practice," he said.

AWS also shared its partner strategy for this year at the briefing. The company said AWS Korea's 2026 partner strategy can be summarized as three points: expanding partner coverage and deeper collaboration, accelerating marketplace growth, and strengthening industry-specific partner strategies.

Hee-ran Bang (방희란), head of partners at AWS Korea, stressed that this is aimed at accelerating growth for both partners and customers by combining industry domain expertise and AI capabilities in an AI era where cloud infrastructure is no longer a differentiator but a basic foundation.