Amazon Bedrock AgentCore Harness Goes GA, Deploying Full Production Agents in Two API Calls

InfraTop News5 sources·Jun 8

agents developer-platforms coding-tools amazon bedrock-agentcore

Summary

• AgentCore harness is now generally available: two API calls (CreateHarness, InvokeHarness) deploy a full production agent in minutes with no orchestration code required
• AgentCore Runtime hosts coding agents in isolated Linux microVMs with identity, MCP gateway, and CloudWatch observability built in — eliminating the open-laptop credential exposure problem
• Three knowledge layers give agents access to organizational data (SharePoint/Drive/Confluence), live web search, and paid data sources with no custom pipelines to build
• Production monitoring and memory capabilities enable teams to diagnose deployed agent failures and improve agents continuously from every interaction
• Governance controls graduate from 'learn mode' to autonomous enforcement as agent confidence grows, giving organizations a safe, graduated path to trusting agents at scale

Adjust signal

Details

Type	Key Point	Context
Product Launch	AgentCore Harness Goes GA: Two-API Production Deployment	CreateHarness + InvokeHarness reduce full production agent setup to two API calls; harness wires Runtime, Memory, Gateway, Browser, Identity, and Observability automatically — in preview since April 2026, now GA June 18, 2026
Tech Info	Any Model, Mid-Session Provider Switching	Harness supports any model provider and allows switching model providers mid-session without losing agent context or accumulated state
Product Launch	AgentCore Runtime: Secure Cloud Hosting for Coding Agents	Isolated Linux microVMs with persistent workspaces let coding agents run off developer laptops, eliminating credential-exposure risk and the operational 'open laptop' problem
Security Alert	Laptop Execution Exposes Developer Credentials to Agents	Running agents locally exposes .env files, ~/.aws/credentials, SSH keys, and private registry tokens via shared shell — a critical least-privilege violation that microVMs solve
New Tech	Three Knowledge Layers: Organizational, Web, and Paid Data	Managed Knowledge Base handles enterprise RAG from SharePoint/Drive/Confluence/S3; Web Search provides live information; a third layer breaks through paid data paywalls for financial and market data — all with no pipelines to build
Infrastructure	Unified MCP Gateway Centralizes Agent Tool Access	Single Model Context Protocol endpoint gives agents access to GitHub, Jira, Slack, and other tools with real credentials held securely outside the agent environment
Tech Info	Production Monitoring + Memory for Continuous Improvement	Monitoring tools diagnose what's failing in deployed agents; memory capabilities let agents learn from every interaction and correction, compounding performance improvements after deployment
Strategy	Governance Controls Scale with Agent Autonomy	Security capability starts in 'learn mode' and graduates to autonomous enforcement as confidence builds — giving organizations a controlled, non-disruptive path to trusting agents at scale

Product Launch

AgentCore Harness Goes GA: Two-API Production Deployment

CreateHarness + InvokeHarness reduce full production agent setup to two API calls; harness wires Runtime, Memory, Gateway, Browser, Identity, and Observability automatically — in preview since April 2026, now GA June 18, 2026

Tech Info

Any Model, Mid-Session Provider Switching

Harness supports any model provider and allows switching model providers mid-session without losing agent context or accumulated state

Product Launch

AgentCore Runtime: Secure Cloud Hosting for Coding Agents

Isolated Linux microVMs with persistent workspaces let coding agents run off developer laptops, eliminating credential-exposure risk and the operational 'open laptop' problem

Security Alert

Laptop Execution Exposes Developer Credentials to Agents

Running agents locally exposes .env files, ~/.aws/credentials, SSH keys, and private registry tokens via shared shell — a critical least-privilege violation that microVMs solve

New Tech

Three Knowledge Layers: Organizational, Web, and Paid Data

Managed Knowledge Base handles enterprise RAG from SharePoint/Drive/Confluence/S3; Web Search provides live information; a third layer breaks through paid data paywalls for financial and market data — all with no pipelines to build

Infrastructure

Unified MCP Gateway Centralizes Agent Tool Access

Single Model Context Protocol endpoint gives agents access to GitHub, Jira, Slack, and other tools with real credentials held securely outside the agent environment

Tech Info

Production Monitoring + Memory for Continuous Improvement

Monitoring tools diagnose what's failing in deployed agents; memory capabilities let agents learn from every interaction and correction, compounding performance improvements after deployment

Strategy

Governance Controls Scale with Agent Autonomy

Security capability starts in 'learn mode' and graduates to autonomous enforcement as confidence builds — giving organizations a controlled, non-disruptive path to trusting agents at scale

Key capabilities and context for Amazon Bedrock AgentCore harness GA and full agent platform

What This Means

Amazon Bedrock AgentCore has reached full GA with the harness abstraction layer making it possible to deploy a production-ready agent in minutes using just two API calls. This completes AWS's vision of making agent infrastructure a configuration problem rather than a build problem — abstracting away sandboxed compute, memory, tool access, web browsing, governance, and observability into a single managed layer. Combined with the platform's knowledge layers, monitoring, and governance controls, AgentCore now represents one of the most comprehensive managed agent infrastructure stacks available, positioning AWS to capture the growing wave of organizations moving AI agents from prototypes into production at scale.

Sentiment

Positive and strategic, with AWS insiders and practitioners highlighting enterprise readiness and new capabilities

@SwamiSivasubramSwami Sivasubramanian · Head of AI Services and Data at AWSView post

Supportive

“Building an agent is simple. Getting it to production is much harder. That’s why we built Bedrock AgentCore, our platform for when you want to move agents from proofs of concept to production. In the last six months, agent workloads have grown 15x. Today we announced managed knowledge bases, web search, new optimization capabilities, security integrations, and GA for AgentCore harness.”

@ElizabethFue12Elizabeth Fuentes · AWS Developer AdvocateView post

Excited

“Bedrock AgentCore now has native web search. Agents get current web data + verified facts from Amazon's knowledge graph, all without leaving your AWS environment. No external APIs.”

@ZeekTytZeek | Insights 📊 · Social Media Strategist & Media ConsultantView post

Positive

“The AWS angle is the real strategic win here: Bedrock AgentCore is already AWS’s framework for building and running these agents. By embedding deep security, monitoring, governance, and recovery directly into the platform, AWS makes Bedrock the default choice for any serious enterprise that wants to deploy agents at scale without taking on unacceptable risk.”

@pricepertokenpricepertoken · LLM News, Model Releases and Pricing UpdatesView post

Informative

“New capabilities in Amazon Bedrock AgentCore enable agents with connections to organizational, web, and paid knowledge plus continuous learning.”

Split

~80/20 positive/neutral with no notable criticism; focus on capabilities vs. pure hype.

Sources

Update history (2)

Jun 18AWS ML Blog confirms AgentCore harness is now generally available (was in preview since April 2026). New article adds: (1) harness GA milestone with CreateHarness + InvokeHarness two-API-call interface; (2) specific capabilities at GA — sandboxed filesystem/shell environment, session-spanning memory, AWS-curated skills catalog, web browsing, MCP tool integration, mid-session model provider switching, real-time streaming with auto CloudWatch tracing; (3) no orchestration code or container building required. Event title and tier1/tier2/tier3/key_facts updated to reflect harness GA and its capabilities alongside existing AgentCore platform coverage.

Jun 17New AWS ML Blog article (June 17, 2026) reveals Amazon Bedrock AgentCore has expanded well beyond its original coding agent hosting scope. Three materially new capability areas added: (1) Three knowledge layers — Managed Knowledge Base for agentic RAG from SharePoint/Drive/Confluence/S3, Web Search for real-time information, and a paid data access layer for financial/market data behind paywalls — all requiring no custom pipelines. (2) Production monitoring and memory capabilities enabling teams to diagnose deployed agent failures and continuously improve agents from every interaction. (3) Governance controls that graduate from 'learn mode' to autonomous enforcement as agent confidence grows. Event title, all tiers, what_this_means, key_facts, and importance_score (7→8) updated to reflect AgentCore's evolution into a full agent lifecycle platform. is_top_news elevated to true.

Similar Events

Amazon Bedrock Adds Open-Weight Model Support for AI Coding Agents via OpenCode Integration

Sep 23

AWS Launches AgentCore Gateway to Centralize AI Agent Tool Access and Security Governance

Aug 21