AWS Taking advantage Automatic reasoningwhich uses math-based validation to develop new capabilities in its Amazon Bedrock Agent Core Companies like Platforms dig deep into the agentic AI ecosystem.
Announced during its annual RE: Invent conference in Las Vegas, AWS is adding three new capabilities to Agent Core: "policy," "Assessment" And "Episodic memory." The new features aim to give businesses more control over agent behavior and performance.
AWS also revealed that it is introducing “a new class of agents," or "frontier agent," which are independent, scalable and independent.
AWS VP for Agentic AI, Swami Sivasubramaniai, told VentureBeat that many of AWS’s new features represent a shift to being a builder.
“We’re actually in the midst of a major tectonic shift with AI, but agentic AI is really starting to change what the art of what’s possible is, and that’s going to make it one of the really transformative technologies,” Sivasubramaniai said.
Policy Agent
New POllie’s ability helps businesses reinforce guidelines even when the agent has already reasoned out its response.
David Richardson, AWS VP for Agent Core, told VentureBeat that the policy tool sits between the agent and its tools, rather than being baked into the agent, as fine-tuning often occurs. The idea is to prevent an agent from violating enterprise rules and redirect it to reevaluate its reasoning.
Richardson gives the example of a customer service agent: A company will write a policy that says the agent can give a refund of up to $100, but for anything else, the agent will need to humanize the customer. They noted that it is easy to corrupt an agent’s reasoning loop by injecting or poisoning data immediately, for example, causing the agent to bypass Guardrail.
“There are always these immediate injection attacks where people try to override the agent’s reasoning so that the agent doesn’t do things it shouldn’t,” Richardson said. “That’s why we implemented an agent-out policy, and it works by using automated reasoning capabilities that we’ve spent years building to help customers define their capabilities.”
AWS unveiled Automated Reasoning Checks On last year’s re-bedrock: invention. These use neurosymbolic AI, or math-based validation, to prove accuracy. The tool applied mathematical proofs to the models to verify that it was not falsified. AWS leans heavily into neurosymbolic AI and automated reasoning, and drives enterprise-grade safety and security in ways that are different from other AI model providers.
Episodic memories and appraisal
Two other new updates to Agent Core, "Assessment" And "episodic memory," Also give businesses a better view of agent performance and give agents episodic memory.
an An extension of agent core memory, episodic memory refers to knowledge that agents tap into only occasionally, in contrast to long-running preferences, which they must constantly refer to. The context window hinders some agents, so they sometimes forget information or conversations they haven’t tapped into for a while.
“The idea is to help capture information that the user really wants the agent to remember when they come back," Richardson said. "For example, ‘What is their preferred seat on the plane for family trips?’ or ‘How is the price range range?’"
Episodic memory differs from previously described agent core memory because, instead of relying on short- and long-term memory retention, agents built on the core can recall certain information based on stimuli. This may eliminate the need for customs instructions.
With Agent Core Assessment, organizations can use 13 pre-built assessors or write their own. Developers can set alerts to alert agents if agents begin to fail quality monitoring.
Frontier Agent
But perhaps AWS’s strongest push into enterprise agent AI is the release of Frontier Agents, or the release of fully automated and independent agents that the company says can act as a bit-directional companion.
The concept is similar, if not identical, to that of competitors’ more passive agents Google And Open Eye. However, it looks like AWS is releasing more than just autonomous coding agents.
Sivasubramaniai told them "New class" of agents, "Not just a step function change you can do today. They move from helping with individual tasks to complex projects."
The first is Kero, an autonomous coding agent that has been in public preview since July. At the time, Kero was billed as an alternative to web coding platforms like OpenAI Codex or Windsurf. Like Codex and Google’s myriad asynchronous coding agents, Including Julescan keycode, perform reviews, independently fix bugs and determine tasks that need to be completed.
AWS Security Agent, meanwhile, embeds deep security expertise into applications from the start. The company said in a press release that users “define security standards once and AWS Security Agent automatically validates them in your requests during its review – helping teams address the risks that matter to their business, not generic checklists.”
The AWS Devops agent will help developers, especially those on call, find system breaks or bugs. It can respond to events using its knowledge of the application or service. It also recognizes the relationship between the application and tools, such as Amazon Cloud Watch, Datadog and Splunk, to trace the root cause of the problem.
Enterprises are interested in deploying agents and ultimately bringing more autonomous agents into their workflows. And, while companies like AWS are empowering these agents with security and control, organizations are slowly figuring out how to connect them all.