I spent this weekend going through the AWS Solutions Library and its customer success stories, hunting for the deployments that have actually shipped. I was specifically interested in AWS agentic AI case studies—not the polished demos, but the ones running in production with real numbers attached to them.
Eleven of them are worth your time. I have put them in a single table below: what each company built, the stack underneath it, and the result they reported. Read it top to bottom and you can watch enterprise AI grow up in front of you, from a security script that quietly patches itself all the way to a team of agents designing hardware for the Moon.
This is the shift I keep coming back to on this blog. For thirty years we told software exactly what to do, and it did exactly that, no more and no less. Instructions in, results out. The strongest cases here work nothing like that. Blue Origin and Cox do not hand their agents a script. They hand them a goal and let them work out the steps. Intent in, outcomes out. It is the same change we are living through on our own regulated origination projects, which is partly why these stories caught my eye.
Source: every case study below is drawn from the AWS Solutions Library — Customer Success Stories (accessed June 2026). All figures and outcomes are as reported by AWS and the named customers.
| Company & Industry | What they built | Core AWS / tech stack | Headline outcome |
|---|---|---|---|
| Blue Origin Aerospace |
BlueGPT, an internal agentic platform built around a secure model gateway, an agent marketplace, and multi-agent orchestration. Specialised agent teams designed TEAREx, a lunar thermal-battery system, running the design loop themselves until it met spec.
|
Amazon Bedrock; Bedrock AgentCore (memory); Bedrock Knowledge Bases; Strands Agents SDK; Amazon EKS; OpenSearch; RDS; Lambda; EC2 P5 & G5 GPUs. | Hardware development time cut by about 90%, from years to days. 2,700+ agents in use at 70% company adoption. |
| Cox Automotive Automotive services & lending |
An enterprise agentic AI program on a reusable reference architecture. Seventeen solutions reached production in under a year, including FleetMate for fleet-repair estimates and VinSolutions for dealer-to-consumer messaging.
|
Bedrock AgentCore (Runtime, Memory, Observability, Identity); Amazon Bedrock; Knowledge Bases; Guardrails; Strands Agents; Titan embeddings; CloudWatch. | Fleet estimates from 8–48 hours down to 30 minutes. 3x consumer response rates. One workflow projected to save 17,000 hours. |
| Mercedes-Benz Automotive manufacturing |
Pulling a sprawling global SAP estate onto AWS for RISE with SAP, then using agentic AI to modernise what is left of the mainframe, including refactoring old COBOL into Java.
|
AWS for RISE with SAP; AWS Transform for mainframe (agentic); EC2 High Memory U7i instances. | OneERP core finance app, used by tens of thousands, migrated in nine months. Targeting up to 60% fewer applications. |
| Pinterest Consumer tech |
Visual discovery at global scale: recommendation models, the homegrown Pinterest Canvas diffusion model, visual search across 2.5B+ objects, and the conversational Pinterest Assistant.
|
Amazon EKS; 10,000+ EC2 G5 (inference); 600+ EC2 P4/P4de (training); SageMaker; Amazon Bedrock. | 10M+ recommendations per second. 70% of discovery now AI-driven. Revenue up 17% YoY. |
| TwelveLabs Video intelligence |
Video-native foundation models, Marengo for embeddings and Pegasus for video-to-text, so you can search and reason over archives running into millions of hours.
|
Amazon Bedrock; EKS; EC2 G6e (L40S) & P5 (H100); S3; S3 Vectors; SageMaker HyperPod. | Petabyte-scale search with sub-second retrieval. S3 Vectors cuts vector-storage cost by up to 90%. |
| Condé Nast Media & publishing |
Moved 800+ properties onto one AWS foundation, built a Databricks lakehouse for cross-brand data, and put generative AI to work on content-rights and moderation across more than 22 brands.
|
EKS; EC2; S3; CloudFront; Lambda; Control Tower; Lake Formation; PrivateLink; KMS; Bedrock (+ Guardrails); SageMaker. Partners: Databricks, Snowplow. | Content-rights work cut from weeks to minutes. The org moved from opinion-led to data-led calls. |
| Phagos Biotech |
Alphagos, a generative platform that matches bacteriophages to target bacteria as an alternative to antibiotics, simulating millions of phage-bacteria interactions.
|
Amazon SageMaker AI; EC2 (GPU); S3; RDS. | A new treatment in roughly two months instead of ten-plus years. 99.5% less screening time. Half the wet-lab tests. |
| AudioShake Audio AI |
Deep-learning source separation that unmixes any recording into clean stems, with separate models for music, dialogue, and overlapping speakers, served by SDK and in real time.
|
EC2 G6 GPUs; S3; ECS & EKS; Lambda + Step Functions; CloudFormation; RDS. | In production for Green Day, Disney Music Group and the Las Vegas Sphere. Won the 2024 re:Invent Unicorn Tank. |
| TUI Group Travel |
A compliance framework on top of the open-source Automated Security Response solution, with playbooks that remediate findings across the estate on their own.
|
Automated Security Response on AWS; Security Hub; GuardDuty. | Up to 156 workdays saved a year. 85% faster remediation. |
| Sony Music Solutions Entertainment |
Brought large-scale load testing in house with the Distributed Load Testing solution, replacing a slow third-party service across hundreds of sites.
|
Distributed Load Testing on AWS; EC2; CloudFormation; ECS + Fargate; Apache JMeter. | Peak requests up 170%, from 0.74M to 2M per minute. Cost per test down 90%+. |
| Notorious Studios Gaming |
A serverless game-analytics pipeline on AWS Guidance, feeding near real-time player telemetry into design decisions for Legacy: Steel & Sorcery.
|
Game Analytics Pipeline on AWS; AWS CDK; Glue; Kinesis Firehose; Kinesis Data Analytics; QuickSight. | Customised in two days, deployed in under ten minutes. Replaced a three-day manual process. |
Where they land on the Agentic AI Roadmap
Once I had the list, I did what we do with every batch of cases on this site. I plotted them against the five levels of the Agentic AI Roadmap, business function up the side, maturity across the bottom.
The shape of it says more than I can. The three hollow dots on the left are not really agentic at all. TUI’s security automation, Sony’s load testing, Notorious’s analytics pipeline: good engineering, no agents. They sit at Prompted because that is where solid classical cloud work lives.
The middle fills up quickly. Phagos, TwelveLabs and AudioShake have generative models in production but not much ceremony around them. Mercedes, Pinterest and Condé Nast add the part most teams skip until something breaks, which is governance. Guardrails, data controls, the unglamorous scaffolding that lets you sleep at night.
Two cases pull away from the pack. Cox Automotive reaches Assured because it built the things you actually need before trusting an agent with anything that matters: observability on every interaction, scoped identity, guardrails, a reference architecture other teams can pick up and reuse. Blue Origin reaches Autonomous because its agents do not just make suggestions. They run the design loop themselves until the hardware passes.
That last one is a bigger deal than it looks. When we mapped fifty enterprise cases a couple of weeks ago, nothing reached Level 5. This smaller, AWS-only set has one that does. I do not think that is noise. The frontier moved this year, and it moved in exactly the order the roadmap predicts. Assurance first, autonomy after.
If you are somewhere in that crowded middle, and most of us are, the useful question is not how to leap straight to Blue Origin. It is which single control you are missing that keeps you one level lower than you would like to be. Find that, build it, and you have your next quarter.
© Dr. Harish Kotadia, 2026. All Rights Reserved.
Dr. Harish Kotadia, Ph.D., is an Enterprise AI Architect with 20+ years of IT consulting experience serving Fortune 100 clients, specializing in agentic AI systems built on Anthropic Claude, AWS Bedrock, and Google Vertex AI. He holds a Ph.D. in Marketing Management with doctoral research in marketing analytics. Follow him on LinkedIn at www.linkedin.com/in/hkotadia and at AgenticAIArch.com.