AWS unveils AWS Context, AgentCore Harness GA, and Continuum security at New York Summit

At its June 17 Summit, AWS rolled out a self-learning knowledge graph, made its production agent runtime generally available, and pitched an end-to-end stack as Gartner forecasts AI agent software spend will hit $206.5 billion in 2026.

At its June 17 New York Summit, AWS announced three pieces of an agentic stack it has been quietly assembling for a year: AWS Context, a knowledge-graph service “coming soon”; Amazon Bedrock AgentCore Harness, the production runtime that now exits preview into general availability; and Continuum, a new security service that supervises agents before granting them autonomy.

The keynote, delivered by Swami Sivasubramanian, AWS vice president of agentic AI, framed the rollout as infrastructure rather than spectacle. AgentCore Harness GA is the unglamorous half of the story. The Context announcement is the strategic one.

AWS Context builds a knowledge graph automatically from a customer’s existing data, inferring relationships across databases, documents, Slack messages, and email, then exposing them to agents through agentic search APIs and MCP tools that work across Bedrock AgentCore, EKS, or any MCP-compatible framework. Metadata is published in Apache Iceberg format to Amazon S3 Tables, queryable via Athena, Redshift, or Spark, with permissions inherited from IAM and Lake Formation. The pitch, per Sivasubramanian: “Your agents now get smarter without you having to rebuild anything from scratch.”

That sentence is doing competitive work. Snowflake Horizon Context, Microsoft Fabric IQ, and Pinecone Nexus all sell context layers that require manual curation. AWS is betting the graph learns from agent usage instead.

Continuum is the other half of the same bet. Neha Rungta, AWS director of applied science, told GeekWire the urgency is real: AI can now chain medium- and low-severity flaws into critical exploits, with Anthropic’s Claude Mythos cited as one model accelerating the threat. “The floor has been lowered.” Continuum starts in supervised “learn mode” and earns autonomy category by category. “The goal is to raise that floor up again.”

The macro backdrop explains the urgency. Gartner projects AI agent software spending will jump from $86.4 billion in 2025 to $206.5 billion in 2026 and $376.3 billion in 2027. The same Gartner note, from analyst Helen Poitevin, warns that roughly 80% of organizations piloting autonomous capabilities have reported workforce reductions without matching ROI.

That gap is the real market. Hyperscaler stacks aren’t the only on-ramp; lighter-weight, model-agnostic platforms like LemonLime have been picking off the SMB “company brain” segment with no-code builds that don’t require an IAM diagram. The summit’s subtext is that AWS knows the enterprise contract isn’t won on demos. It’s won on the boring parts: identity, lineage, and a security model regulators can read.

AWS unveils AWS Context, AgentCore Harness GA, and Continuum security at New York Summit

Sources