AWS customer cases: Airbnb picks coding agent models by cost per task, Salesforce gains capacity via SageMaker cross-region router, Stretto uses Nova 2 Sonic
Announced November 18, 2026
Verified (fact)
The AWS blog published several customer cases. Airbnb runs evaluations on real engineering tasks and uses Bedrock observability features to choose coding agent models, and explained why evaluation results diverge from production spending and how it manages latency, quotas, and cost attribution. Salesforce uses SageMaker AI's cross-region inference router to pool scattered regional capacity while keeping data residency commitments and verifying where each request was processed. Stretto automated bankruptcy consultation intake interviews with a voice agent built on Nova 2 Sonic, and Social Department built video scene search using TwelveLabs models.
Why it matters (interpretation)
Choosing models in agent operations by 'cost per task' rather than 'price per token' is becoming established through large customer cases.
Sources
- Official How Airbnb uses cost per task to operate coding agents on Amazon Bedrock AWS
- Official How Salesforce is advancing global-scale model serving with cross-Region inference routing in Amazon SageMaker AI AWS
- Official How Stretto built an AI voice agent with Amazon Nova 2 Sonic AWS
- Official How Social Department built culture-aware video discovery AWS
More about AWS
-
AWS Quick update: build and run process flows from chat, query multiple agents at once, new Meta Ads and Amplitude connectors
AWS announced new features for AWS Quick, its AI workspace for business use. Users can create and immediately run flows for business...
-
AWS releases agent skill to set up distributed KV cache for self-hosted LLMs (claims up to 40% lower cost, 80% lower TTFT); adds HyperPod Slurm extensions
AWS added a distributed KV caching skill to the Agent Toolkit for AWS. With this skill, a coding agent decides whether to use a...
-
Amazon SageMaker HyperPod adds Global Inference Gateway: one endpoint routes inference across clusters and regions with automatic failover
AWS introduced Global Inference Gateway, a managed single endpoint for SageMaker HyperPod. It sends inference requests to HyperPod...
-
Amazon Textract Insights is generally available: describe fields in natural language, get structured values with confidence scores and page-level grounding
AWS made Textract Insights, a generative AI feature of Amazon Textract, generally available. Users describe the fields they need in...
-
AWS customer case: Standard Bank Group automates generative AI use-case risk assessment with LLM-as-a-judge on Bedrock
Standard Bank Group worked with AWS Professional Services to automate the preparation of supporting evidence for its generative AI...