AWS Rebuilds AgentCore Runtime to Cut AI Agent Costs and Cold-Start Delays
The updated runtime is designed for production agents that can run for extended periods, operate without supervision and handle workloads triggered by events rather than direct user requests.
Amazon Web Services (AWS) has introduced a new version of its Amazon Bedrock AgentCore runtime, targeting two growing challenges for enterprises running AI agents at scale– unpredictable startup times and infrastructure costs.
The updated runtime is designed for production agents that can run for extended periods, operate without supervision and handle workloads triggered by events rather than direct user requests. AWS said the platform has been used by thousands of teams since its launch.
The Left Shift Bureau
The new runtime changes how memory is allocated and reclaimed during an agent session. Instead of maintaining the highest amount of memory allocated throughout the session, the platform can reclaim memory when it becomes inactive and allocate additional resources as workloads require them. This allows billing to more closely follow actual resource consumption.
"It brings better memory management, reclaiming memory as a session releases it instead of holding it at the peak. It also delivers consistent cold start times regardless of container size or agent concurrency. You get the serverless model you already liked, now more elastic. Memory is released back the instant a session ends, startup times stay consistent regardless of size or concurrency, and the bill tracks the work your agent does," AWS said.
According to AWS testing, the new runtime recorded a P75 cold-start latency of around two seconds for container images ranging from 200 MB to 2 GB. The previous runtime's latency increased with image size, ranging from roughly 5.4 seconds to nearly 30 seconds in the company's tests.
The Left Shift Bureau
The improvement comes from preparing an agent environment once, capturing a snapshot and restoring that snapshot for new instances. AWS said the approach avoids repeating the boot and initialisation process each time an agent starts.
The runtime continues to support scale-to-zero and consumption-based billing, allowing customers to avoid paying for idle capacity. AWS said the new model charges based on memory actually used rather than keeping an entire container footprint active.
AWS also outlined upcoming additions including baseline pricing for consistently active workloads, larger compute and storage options, x86 support, greater session lifecycle controls and scoped identities for unattended agents.
The latter could become particularly important as enterprises deploy agents capable of acting without continuous human supervision, allowing each session to operate with defined permissions.
The Left Shift Bureau
The Left Shift Bureau
The Left Shift Bureau
