The near-term future of enterprise AI will be model-driven, democratized, powered by agents, and under increasing pressure to control costs.

These key trends were highlighted in today’s keynote presentation by Swami Sivasubramanian (pictured), vice president of AI and data at Amazon Web Services Inc., who spoke at the company’s annual re:Invent gathering in Las Vegas. His remarks followed a series of AWS announcements on Tuesday that defined the cloud giant’s approach to enterprise AI and model usage. Sivasubramanian made it clear that his company remains closely attuned to the fast-moving dynamics that surround the evolving world of enterprise AI.

“It’s been a monumental year,” Sivasubramanian said. “Disruption is the new normal.”

The announcements today showed how AWS is seeking to keep pace with developments in the explosive growth of AI model adoption. In September, Hugging Face Inc., the open-source hub for AI models, announced that it surpassed a million models in its repository, a near-doubling from what it reported in April.

As enterprise AI adoption expands, AWS has focused on supplying tools for enterprise clients to manage data-fueled models used for making decisions and predictions. Wednesday’s news included the addition of new models and providers within Amazon Bedrock, along with capabilities for Amazon SageMaker to build and deploy machine learning models for any use case within fully managed infrastructure.

“We are currently facing an inflection point when it comes to model training,” Sivasubramanian said. “We offer a diverse set of options that are capable of tackling any task imaginable.”

Prompt caching and AI agents

Amazon Bedrock is also receiving enhancements designed to help enterprises control costs associated with increased AI deployment. AWS announced new prompt caching tools for Bedrock that will reduce repeated processing which can run up a sizable bill. The company also unveiled Intelligent Prompt Routing that will automatically direct prompts to different foundation models and seek affordable alternatives.

“Bedrock will automatically route your prompt to the model that will give you the best response at the lower cost,” Sivasubramanian said.

The enhancements to Bedrock also highlighted enterprise interest in deploying and managing AI agents. AWS customers such as Argo Labs are using Intelligent Prompt Routing for its voice agent solutions used by restaurants. When diners call to place orders or book tables, the chatbot can dynamically route the query to the most suitable model for a response.

“Agents unlock new levels of automation that were not possible before,” Sivasubramanian told the re:Invent gathering.

The impact of AI agents is also becoming more visible in the world of home mortgage lending. In a presentation during the keynote session on Wednesday, Shawn Malhotra, chief technology officer of the fintech platform Rocket Companies Inc., described how his firm used AI agents to guide clients through the mortgage application process. He claimed that it has become three times more likely that Rocket will close a client via an AI-driven chat than through a human interface.

Malhotra, who joined Rocket in May after leading AI initiatives for Thomson Reuters, painted a picture of AI agents as an important step for transforming the financial world. “The journey to own a home is still riddled with friction and stress,” Malhotra said. “This is an industry begging to be disrupted.”

Democratizing machine learning

AI’s role as a catalyst for disruption has been a continuing theme at re:Invent this week. In an exclusive interview with SiliconANGLE in advance of the conference, Sivasubramanian spotlighted generative AI’s ability to democratize data for nontechnical users as one of the technology’s more significant achievements.

The AWS executive, who has been with the cloud giant for 19 years, led the company’s development of the DynamoDB database for running high performance applications at scale, and has been instrumental in building AI and machine learning tools as product offerings for the firm. One of those became SageMaker, launched at re:Invent in 2017 as a unified platform for data, analytics and AI. On Wednesday, Sivasubramanian announced that Amazon Q Developer would now be available in SageMaker Canvas, enabling users to connect machine learning expertise with business needs.

By making machine learning more accessible through a natural language interface, capabilities once limited to data scientists and AI experts will now become readily available to nontechnical users. AWS views Q Developer as a key step in democratizing data for business users.

“With Q, you can ask for insights in natural language and get dashboards or data stories in minutes,” Sivasubramanian told SiliconANGLE. “Tasks that used to take weeks now happen in seconds.”

Photo: Robert Hof/SiliconANGLE