AWS is leveraging automated reasoning, which uses math-based verification, to construct out recent capabilities in its Amazon Bedrock AgentCore platform as the corporate digs deeper into the agentic AI ecosystem.
Announced during its annual re: Invent conference in Las Vegas, AWS is adding three recent capabilities to AgentCore: "policy," "evaluations" and "episodic memory." The brand new features aim to present enterprises more control over agent behavior and performance.
AWS also revealed what it calls “a brand new class of agents," or "frontier agents," which might be autonomous, scalable and independent.
Swami Sivasubramanian, AWS VP for Agentic AI, told VentureBeat that lots of AWS’s recent features represent a shift in who becomes a builder.
“We are literally on the cusp of a serious tectonic transformation with AI, but agentic AI is really starting to rework what’s the art of the possible, and it will make this one of the crucial truly transforming technologies,” Sivasubramanian said.
Policy agents
The brand new policy capability helps enterprises reinforce guidelines even after the agent has already reasoned its response.
AWS VP for AgentCore David Richardson told VentureBeat that the policy tool sits between the agent and the tools it calls, somewhat than being baked into the agent, as fine-tuning often is. The concept is to stop an agent from violating enterprise rules and redirect it to re-evaluate its reasoning.
Richardson gave the instance of a customer support agent: An organization would write a policy stating that the agent can grant a refund of as much as $100, but for anything higher, the agent would wish to bounce the client to a human. He noted that it stays easy to subvert an agent's reasoning loop through, for example, prompt injection or poisoned data, leading agents to disregard guardrails.
“There are at all times these prompt injection attacks where people attempt to subvert the reasoning of the agent to get the agent to do things it shouldn’t do,” Richardson said. “That’s why we implemented the policy outside of the agent, and it really works using the automated reasoning capabilities that we’ve spent years increase to assist customer define their capabilities.”
AWS unveiled Automated Reasoning Checks on Bedrock ultimately 12 months’s re: Invent. These use neurosymbolic AI, or math-based validation, to prove correctness. The tool applies mathematical proofs to models to substantiate that it hasn’t hallucinated. AWS has been leaning heavily into neurosymbolic AI and automatic reasoning, pushing for enterprise-grade safety and security in ways in which differ from other AI model providers.
Episodic memories and evaluations
The 2 other recent updates to AgentCore, "evaluations" and "episodic memory," also give enterprises a greater view of agent performance and provides agents episodic memory.
An enhancement of AgentCore memory, episodic memory refers to knowledge that agents tap into only occasionally, unlike longer-running preferences, which they must refer back to consistently. Context window limits hamper some agents, so they often forget information or conversations they haven’t tapped into for some time.
“The concept is to assist capture information that a user really would need the agent remembered after they got here back," said Richardson. "For instance, 'what’s their preferred seat on an airplane for family trips?' Or 'what’s the type of price range they're on the lookout for?'"
Episodic memory differs from the previously shipped AgentCore memory because, as a substitute of counting on maintaining short- and long-term memory, agents built on AgentCore can recall certain information based on triggers. This may eliminate the necessity for custom instructions.
With AgentCore evaluations, organizations can use 13 pre-built evaluators or write their very own. Developers can set alerts to warn them if agents begin to fail quality monitoring.
Frontier agents
But perhaps AWS's strongest push into enterprise agentic AI is the discharge of frontier agents, or fully automated and independent agents that the corporate says can act as teammates with little direction.
The concept is analogous, if not an identical, to those of more asynchronous agents from competitors like Google and OpenAI. Nonetheless, AWS appears to be releasing greater than just autonomous coding agents.
Sivasubramanian called them a "recent class" of agents, "not only a step function change in what you possibly can do today; they move from assisting with individual tasks to complex projects."
The primary is Kiro, an autonomous coding agent that has been in public preview since July. On the time, Kiro was billed as a substitute for vibe coding platforms like OpenAI’s Codex or Windsurf. Much like Codex and Google’s myriad asynchronous coding agents, including Jules, Kiro can code, undertake reviews, fix bugs independently and determine the tasks it needs to perform.
AWS security agent, meanwhile, embeds deep security expertise into applications from the beginning. The corporate said in a press release that users “define security standards once and AWS security agent mechanically validates them across your applications 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, proactively find system breaks or bugs. It could actually reply to incidents using its knowledge of the applying or service. It also acknowledges the relationships between the applying and the tools it taps, comparable to Amazon CloudWatch, Datadog and Splunk, to trace the basis explanation for the problem.
Enterprises are interested by deploying agents and, eventually, bringing more autonomous agents into their workflows. And, while corporations like AWS proceed to bolster these agents with security and control, organizations are slowly determining methods to connect all of them.
