Amazon Web Services (AWS) chief executive Matt Garmin projected a full agentic artificial intelligence (AI) future for the cloud infrastructure supplier’s customers during his keynote speech at its Re:Invent show in Las Vegas.
He told attendees that freedom for customers to invent has been the consistent mantra of AWS since its inception two decades ago, and that the true business value of artificial intelligence will be unlocked with the increased deployment of AI agents.
Garmin said AWS has been innovating at “all levels of the technology stack” during the past three years, and that “billions of agents” generating real-world value is the supplier’s vision.
He was joined onstage by executives from AWS customers Sony and Adobe.
John Kodera, chief digital officer at Sony, talked about the Japanese value of “Kando”, meaning deep emotional engagement and awe as a business goal. He referred to Sony’s “data ocean”, built on AWS’s machine learning platform Bedrock, which is used to build generative artificial intelligence (GenAI) applications. Kodera also disclosed that Sony is a user of AWS Nova Forge, a service that enables customers to build so-called “frontier models” using the supplier’s cluster of foundation models, Nova, available on Amazon Bedrock. By a “frontier model”, the industry tends to mean AI systems that are large-scale and often multi-modal, processing data from images, video and audio as well as text.
Shantanu Narayen, chair and CEO of Adobe, also appeared onstage as an AWS customer, with creativity in the era of AI his theme. Adobe software like Acrobat, Firefly and Studio use such AWS technologies as Amazon SageMaker and Bedrock, he said.
AWS’s Garmin said AI startups were “flocking to AWS”, and cited as one AudioShake, a sound separation specialist provider that can help people with hearing difficulties.
Against technical debt
Included among the product advancements the supplier announced at the conference were agentic capabilities in AWS Transform, said to enable rapid modernisation of any code or application.
AWS Transform can be used for “full-stack Windows modernisation, to offload complex, tedious modernisation work across the Windows application stack”.
“You can now identify application and database dependencies, and modernise them in an orchestrated way through a centralised experience,” the supplier said.
Air Canada, Experian, QAD, Teamfront, Thomson Reuters and Verisk are using AWS Transform to help eliminate their technical debt, said AWS.
The supplier also announced AWS AI Factories, said to transform customers’ existing infrastructure into high-performance AI environments.
These “factories” consist of dedicated AWS AI infrastructure deployed in customers’ existing datacentres. They are said to combine Nvidia accelerated computing platforms with AWS Trainium chips, high-speed, low-latency networking, and AWS AI services.
Customers can use their existing datacentre space, network connectivity and power while AWS handles deployment and management of the integrated infrastructure. They can enable customers to meet their data sovereignty and regulatory requirements, with accelerated deployment timelines, said AWS.
IDC analyst perspective
Neil Ward Dutton, an IDC analyst focused on AI, automation, data and analytics, said three announcements on the first day of the conference were of note. “AWS AI Factories is one,” he said. “These are AWS AI infrastructure and software stacks, built and managed by AWS, but deployed in customers’ own datacentres. These are, at least initially, aimed at the largest customers with the most challenging security or sovereignty requirements.”
The second, he said, was Amazon Nova Forge. “This is the ability for customers to bring their own training data corpuses to part-trained Nova models, completing model training to create domain-specialised models that can be deployed to Bedrock,” said Ward Dutton.
The third is the “policy specification and evaluations in Bedrock AgentCore”, which is a group of agent-based AI tools and products announced by AWS in July 2025. He said the new elements “specify verifiable operational policies that constrain AI agents’ access to tools, resources and data”.
“The first wave of enterprise GenAI activity is primarily focused on accessing ‘AI at a distance’,” added Ward Dutton. “This only takes you so far, and organisations are now in the middle of finding that out. What will get enterprises to true value at scale is not what got them to this point.
“The next phase of enterprise AI activity – if it is to be successful – has to embrace the absorption and integration of AI technologies into business processes, workflows, systems of engagement, data and knowledge, security and privacy approaches, regulatory responses, and geopolitical and sovereignty concerns,” he said.