Amazon Web Services (AWS) leverages its AI agent platform, Bedrock Agent, to make building and monitoring AI Agents easier for enterprises.
AWS announced a slew of new agent features on Tuesday during the company’s annual event AWS RE: Negon ConferenceSee rank-. The company announced new tools to manage the Agent’s AI limits, Agent memory capabilities, and Agent evaluation features.
One upgrade is the introduction of policies in agentcore. This feature allows users to set boundaries for agent interactions using natural language. This restriction integrates with the agent gateway, which connects the AI agent with external tools, automatically checking the actions and stopping those that may violate the written controls.
Policies allow developers to set access controls to internal data or third-party applications such as Salesforce or slack. The limit can also tell an AI agent that can handle a refund up to $ 100 but must bring a human in the loop of anything bigger, the Vice President of AgentCore, told TechCrunch, told TechCrunch, told techcrunch.
The company also announced agentcore evaluation, which is a suite of 13 pre-built evaluation systems for AI agents that monitor factors including correctness, safety, and accuracy selection tools, among others. It also allows developers to build their own evaluation features.
“That’s really going to help address the biggest fear that (the) agencies have,” Richardson said of the new evaluation capability. “(It’s) something that a lot of people want, but can’t build.”
AWS also announced that it is building in-memory capabilities into its agent platform, agencore memory. This feature allows agents to develop a log of information about users over time, like flight times or hotel preferences, and use that information to inform them about the future.
TechCrunch events
San Francisco
I’m fat
October 13-15, 2026
“Across those three things, we continue to be different layers in agentcore,” Richardson said. “Talking to existing systems with policies, (making agents) more powerful with (AgentCore Memory), helping the development team with agents.”
When that agent Afternoon day Of the current AI industry, some believe the technology will not last. But Richardson thinks the growing toolkit can withstand a fast-paced market despite changing trends — which is expected.
“Being able to take advantage of the reasoning capabilities of the model, coupled with being able to do real things through the tool, sense sustainable patterns,” Richardson said. “The way the pattern works is going to change. I think we’re ready.”
Follow along with all of TechCrunch’s coverage from The Annual Tech Enterprise Event is hereSee rank-.