AWS developing high-performing autonomous AI agents
The cloud giant is deeply invested in generative AI. It has focused on adapting the Amazon Q assistant for what it calls 'high-performing agentic enterprise applications.'
As the biggest generative AI players fight it out with consumer-facing chatbots while also trying to crack the business market, cloud leader AWS is pursuing a lower-profile but equally ambitious strategy in the mushrooming arena of agentic AI, directed nearly exclusively at enterprises.
AWS' AI agents are the offspring of Amazon Q, the tech giant's all-purpose generative AI assistant. The vendor has adapted Amazon Q on its Bedrock generative AI platform for myriad uses, notably to automate tasks such as reviewing and approving code and sifting through and retrieving relevant information from massive enterprise databases.
"If you just look at the industry right now, software development is an area where agentic AI makes a ton of sense. Everyone has a ton of software tasks," said Deepak Singh, an AWS vice president who leads the vendor's Developer Agents and Experience organization, on the Targeting AI podcast from Informa TechTarget.
The Amazon Q code developer agent "essentially acts as the first code reviewer," Singh said. "Do you have code that could potentially be unsafe? Are there any vulnerabilities in this code? And it flags all of them."
AWS often highlights Bedrock's safety guardrail-building capabilities, which also extend to agents. Ensuring that agents do what enterprises want them to do -- reliably, accurately and safely -- is a hot topic, just as trustworthy generative AI has become one of the biggest issues in the larger AI industry.
"We added this automated reasoning capability that verifies that your answers are factually correct into Bedrock," Singh said. "It uses automated reasoning to verify that what the AI is doing is verifiably correct or not. I can build agents that use all those tools to give me a sense of confidence that the results at the other end are going to be valid results, useful and safe.
"In software development, you add a human in the loop. In other use cases, the way you build it, you are adding a bunch of checks and balances in the way. And quite honestly, for most of our customers, they're still adding a human in the loop in many cases, just to make sure," he said. "And as their comfort level improves, they'll slowly dial back on the human in the loop side because they are more comfortable with the tools."
Shaun Sutner is senior news director for Informa TechTarget's information management team, driving coverage of artificial intelligence, unified communications, analytics and data management technologies. Esther Shittu is an Informa TechTarget news writer and podcast host covering artificial intelligence software and systems. Together, they host the Targeting AI podcast series.
