AI agents are expensive to run: they sit idle waiting for a user, a tool or another agent, and the meter keeps running. This week's AWS roundup, published on 9 November 2026, includes several launches that attack that problem — faster starts, less idle waste, and less plumbing to build yourself.
These are launches and updates from the past week, gathered in the official AWS News Blog weekly roundup and cross-checked against AWS AgentCore release notes. They were not all announced on 9 November itself. And one item needs a careful read on price — the cheaper-looking option is not cheaper per unit.
1. AgentCore Runtime V2 reaches 10 regions
AgentCore Runtime V2 is now available in five more regions — Frankfurt, Stockholm, Mumbai, Singapore and Sydney — taking V2 to 10 AWS Regions.
Here is what is actually new. V2 itself became generally available on 18 September 2026 — the November news is the expansion to 10 regions, not the launch. V2 starts your agent from a snapshot, so cold starts stay fast and consistent regardless of concurrency or image size. V2 agents can also now hold up to 250 concurrent connections per session with the default 2-vCPU allocation, applied automatically. A slow cold start means you pay for waiting, or you pay to keep agents warm — snapshot starts cut that waste.
2. Web Answer on AgentCore Gateway
Web Answer on AgentCore Gateway, now generally available in US East (N. Virginia), us-east-1, gives an agent a single tool call returning a synthesised, cited answer to a natural-language question.
Without it, you build your own search, retrieval and citation pipeline, each step costing money and engineering time. Web Answer replaces that plumbing with one call — the saving is fewer moving parts, not a cheaper token.
3. DynamoDB bulk put items
DynamoDB bulk put items loads line-delimited DynamoDB JSON or Amazon Ion files from Amazon S3 into existing tables as a managed, asynchronous job.
Agents that need data often start by loading large files into a database. A managed bulk job handles the load in the background, without a custom loader to write and keep running.
4. S3 Tables managed migration
The fourth item is S3 Tables managed migration: Parquet tables registered in the AWS Glue Data Catalog move into S3 Tables (Apache Iceberg) without building a pipeline or provisioning compute — start, sync as often as you need, complete.
Agent workloads often live on data already stored as Parquet files. Moving that data into a format agents can query efficiently usually means building a migration pipeline of your own. This launch removes that pipeline.
5. Lambda queues for self-managed Kafka
Lambda now supports queues for Apache Kafka in self-managed Kafka event sources, processing records from a single Kafka partition across many parallel Lambda invocations.
Kafka partitions are normally processed in order, one consumer at a time, which limits how fast a queue can drain. Processing one partition across many parallel Lambda invocations lets the work spread out, so queues clear faster.
The price warning: V2 is not cheaper per unit
Here is the honest part: V2 consumption pricing is higher per unit than V1 — $0.1276 per vCPU-hour and $0.0169 per GB-hour, against $0.0895 and $0.00945 for V1. V2 is not cheaper per unit.
What V2 buys is snapshot starts and idle memory reclaimed after 120 seconds, which can lower the cost of long sessions that mostly wait. A committed baseline of $0.0997/$0.0132 per vCPU-hour/GB-hour was listed as launching by October 2026, narrowing the gap for steady workloads.
The lesson is simple: if your agents run flat-out, V2 costs more per unit. If they wait a lot, the idle-memory reclaim and fast snapshot starts can make the bill smaller. Measure your own pattern before switching.
Also in the roundup: AWS Interconnect last-mile preview with Equinix, a private managed connection from its 282 data centres to AWS.
Key takeaways
- AgentCore Runtime V2 is now in 10 regions (Frankfurt, Stockholm, Mumbai, Singapore, Sydney); V2 itself went generally available on 18 September 2026.
- Web Answer on AgentCore Gateway, generally available in us-east-1, gives agents a single tool call for a synthesised, cited answer.
- DynamoDB bulk put items loads JSON or Ion files from S3 into tables as a managed, asynchronous job.
- S3 Tables managed migration moves Parquet tables from the Glue Data Catalog into Apache Iceberg tables without a pipeline.
- Lambda now processes records from a single self-managed Kafka partition across many parallel invocations.
- Watch the price: V2 costs more per unit than V1; the saving comes from snapshot starts and idle-memory reclaim after 120 seconds.
Conclusion
None of these five launches is dramatic alone. Together, they point in one direction: AWS is chipping away at the hidden costs of running agents — the cold starts, the idle minutes, the data-loading plumbing and the queue bottlenecks. The biggest saving will come from matching the right tool to how your agents actually behave, and from checking the price per unit first.
FAQ
Is AgentCore Runtime V2 new?
No. V2 became generally available on 18 September 2026. The November news is that V2 expanded to five more regions — Frankfurt, Stockholm, Mumbai, Singapore and Sydney — taking it to 10 AWS Regions.
Is V2 cheaper than V1?
Not per unit: V2 is $0.1276 per vCPU-hour and $0.0169 per GB-hour, against $0.0895 and $0.00945 for V1. The saving comes from snapshot starts and idle memory reclaimed after 120 seconds.
How do I opt in to AgentCore Runtime V2?
Set platformVersion to V2 when creating or updating a runtime.
Sources
- https://aws.amazon.com/blogs/aws/aws-weekly-roundup-agentcore-runtime-v2-amazon-dynamodb-bulk-put-items-and-more-november-9-2026/
- https://docs.aws.amazon.com/bedrock-agentcore/latest/devguide/release-notes.html
Related reading on Daily Tech & AI:
Researched with AI assistance and reviewed for accuracy.
Follow on WhatsApp
Follow Daily Tech & AI on WhatsApp for the latest tech and AI updates: https://www.whatsapp.com/channel/0029Vb9XzdHDJ6H4TFNAl61r
Comments
Post a Comment