Key Takeaways
- Amazon Bedrock AgentCore now offers fully managed services for deploying autonomous AI agents across procurement, compliance, and IT support workflows at enterprise scale.
- AgentCore Policy, which reached general availability on March 3, 2026, applies guardrails outside the agent’s reasoning loop — treating agents as autonomous actors whose decisions require verification before reaching tools or data.
- AWS VP of Agentic AI Swami Sivasubramanian called AI agents “the most impactful change we’ve seen in ages” and described a shift toward “service as a software.”
- Gartner forecasts that 40% of enterprise applications will include task-specific AI agents by the end of 2026, with 92% of global companies planning to increase AI investments.
What Happened
Amazon Web Services expanded its Amazon Bedrock AgentCore platform with new capabilities that allow enterprises to deploy autonomous AI agents for procurement, compliance monitoring, IT support, and cross-departmental workflow automation. The platform reached general availability in October 2025, and AWS has steadily added governance and evaluation layers since — most recently AgentCore Evaluations on March 31, 2026.
The distinguishing feature is that these agents operate with minimal human oversight. Rather than waiting for approval at each step, agents plan, reason, and execute multi-step tasks across enterprise systems — from validating supplier compliance documents to triaging IT tickets and routing procurement requests.
Why It Matters
The release marks AWS’s direct move to position AI agents not as assistive chatbots but as autonomous operational infrastructure. Swami Sivasubramanian, VP of Agentic AI at AWS, framed the shift bluntly: “Agents are the most impactful change we’ve seen in ages. With agents comes a shift to service as a software. This is a tectonic change in how software is built, deployed and operated.”
AWS is not alone in this push — Microsoft’s Copilot Studio and Google’s Vertex AI Agent Builder both offer competing agent frameworks — but AgentCore’s emphasis on policy enforcement and session isolation targets a specific enterprise concern: how do you let an AI agent act autonomously while maintaining auditable control?
Technical Details
AgentCore comprises three main layers. First, a serverless runtime with complete session isolation that AWS describes as the longest-running workload available on the platform, designed for agents that need to maintain state across extended multi-step operations. Second, Strands Agents, an open-source SDK that lets developers build agents using any foundation model — whether hosted on Bedrock or elsewhere. Third, Amazon Nova Act for building and managing fleets of agents that automate production UI workflows.
The AgentCore Policy layer, which reached GA on March 3, 2026, operates outside the agent’s reasoning loop. This means policy controls are applied as a verification layer rather than embedded in prompts — the system treats agents as autonomous actors whose decisions must be checked before they reach tools, systems, or data. AgentCore Evaluations, which went GA on March 31, 2026, adds quality scoring for agent outputs before they trigger downstream actions.
For procurement specifically, AWS demonstrated a Smart Procurement Assistant built with AgentCore and Strands Agents that automates supplier validation, compliance document review, and purchase order routing — tasks that previously required manual review across multiple enterprise systems.
Who’s Affected
Enterprise operations teams in procurement, compliance, and IT support are the immediate targets. Organizations already using AgentCore in production include Itaú Unibanco, Cox Automotive, Thomson Reuters, Workday, S&P Global Market Intelligence, and PGA TOUR. Mindflow, an agentic automation platform built on AWS with Amazon Bedrock, connects autonomous agents to over 4,000 integrations.
For developers, the open-source Strands Agents SDK lowers the barrier to building custom agents without locking into a single model provider. For compliance and risk teams, the external policy enforcement model introduces a governance pattern that may become standard across the industry as agent autonomy increases.
What’s Next
Gartner projects that 40% of enterprise applications will embed task-specific AI agents by late 2026, and AWS is positioning AgentCore as the deployment layer for that transition. The immediate question is whether enterprises will trust agents to operate without human-in-the-loop approval for high-stakes operations like procurement spending and regulatory compliance — and whether AWS’s external policy enforcement model provides sufficient safeguards. The next major test comes as competing governance frameworks from Microsoft and Google mature and enterprises begin standardizing their agent deployment patterns.
