How Blue Origin Built the First AI Agent-Designed Hardware for the Moon in Days, Not Years

Building hardware that can survive the lunar night — 14 Earth days of freezing darkness — is a challenge unlike anything on Earth. Blue Origin faced a challenge that extended beyond the technical: aerospace engineering requires deep specialization across dozens of disciplines, and that practical knowledge doesn't live in textbooks or public datasets. It's buried in internal systems, databases, and in expert engineer’s minds.

Early generative AI showed promise, but base large language models weren't trained on the intricacies of industrial manufacturing, thermal battery design, or regolith extraction. And with strict security, trade, and export control requirements, Blue Origin needed AI that could work with proprietary knowledge while maintaining the highest standards of data protection. "The early days of GenAI were exciting for us, but limited to more common sorts of tasks," said Blue Origin VP of Enterprise Technology, Will Brennan. "We needed AI that could actually do the specialized work, not just assist with it."

The traditional approach to developing lunar hardware included years of iterative design cycles, with small teams of specialists working sequentially through requirements, simulations, and prototypes. For a company trying to accelerate humanity's return to the Moon, that timeline was incompatible with their mission.

Blue Origin built an AI ecosystem called BlueGPT — a secure LLM gateway, agent marketplace, and multi-agent orchestration platform built using Strands Agents SDK for a model-based agentic design framework, Amazon Bedrock for agent deployment and model access, Amazon EKS for agent runtime envions, Amazon OpenSearch for Retrieval-Augmented Generation (RAG) knowledge bases, Amazon RDS for general purpose databases, and AWS Lambda for scaling serverless workflow automations across all services. The platform empowers every Blue Origin employee to create and deploy specialized AI agents, providing access to proprietary knowledge bases, custom tools, and leading models. Jay Naves, AWS Sr. Solutions Architect, notes that, “AWS provides Blue Origin with access to cloud specialists knowledgeable in the aerospace domain, GPU accelerated computing designed to scale, and advanced agentic capabilities built to automate so that they can innovate at an ever-improving pace.”

Blue Origin used a variety of AWS services, including:

• Amazon Bedrock for foundation model access and management, Amazon Bedrock Knowledge Bases for leveraging industry-specific semantic knowledge, and Amazon Bedrock AgentCore for both short-term (session) and long-term (persistent) agentic memory capabilities. Most notably, Amazon Bedrock AgentCore provides Blue Origin with automatic insight extraction, a hierarchical memory system for context-aware AI interactions, and built-in security and access controls organized in hierarchical namespaces.
• Amazon Strands Agents SDK for model-driven AI agent orchestration, emphasizing LLM reasoning over rigid workflow control. The architecture follows a hierarchical coordination model with specialized domain agents that adapt their behavior based on query complexity and context. This development framework reduces AI application development time from weeks to hours/days (10x improvement) with reusable agent patterns and scales rocket production capacity without proportional workforce increases through trusted AI assistance.
• Amazon Elastic Kubernetes Service (EKS) for fully automated compute, storage, and networking management of Blue Origin’s Kubernetes clusters. This containerized platform provides agent runtime envisions, extensible MCP servers, data extraction and risk management solutions, custom manufacturing and execution systems, and many other microservice capabilities; all capable of leveraging spot instance types for significant cost-savings.

BlueGPT now has over 2,700 agents created and deployed across the company, driving 3.5 million interactions last month alone, with 70% company-wide adoption. Engineers use agents to write code. Manufacturing teams use them to improve work orders and resolve non-conformances 70% faster. Operations teams communicate design changes with suppliers through AI intermediaries.

Using agents developed on BlueGPT, Blue Origin built the Thermal Energy Advanced Regolith Exchanger (TEAREx), a thermal battery system designed to help lunar systems survive the brutal two-week lunar night by extracting heat energy to and from lunar soil stored in an optimized chamber. With BlueGPT's multi-agent capabilities, Blue Origin's In-Space Systems team assembled a specialized team of AI agents — supervisors, librarians, requirements agents, design agents, and analysis agents — to work alongside 2-3 human engineers.

The breakthrough was automated iterative design loops. Connected to AI-native tools, the agents ran complex physics simulations on GPU-accelerated Amazon EC2 instances from the P5 and G5 instance families, evaluated results against requirements, iterated on the design, and executed until specifications were met — autonomously. Using this agentic approach, overall, Blue Origin was able to:

• Reduce hardware development by 90% — from years to days, concept to printed part
• Accelerate analysis workflows by 6x — tasks that took 4 days now take 4 hours
• Deliver mass-optimized designs through automated topology optimization cycles
• Scale specialized expertise — small teams of 2-3 working with AI agents to deliver the work of dozens
• Create the world's first agentically-designed lunar hardware ready for deployment on the Moon

The agents in BlueGPT aren't one-off solutions either. They live in BlueGPT's marketplace, ready to be reused, recombined, and applied to tubes, barrels, harnesses, assemblies, and eventually entire vehicles.

The development of TEAREx demonstrates that small teams equipped with the right AI agents can move at speeds previously unimaginable. In aerospace, that means they can do so without sacrificing the rigor that spaceflight demands.

"This is what we see as the future of engineering teams at Blue Origin," Brennan noted. "Small teams of 2 or 3 working with large teams of AI agents to deliver the work of dozens, and at orders of magnitude faster than before." The company isn't stopping with TEAREx. They're building an enterprise knowledge graph on AWS to transform AI accessibility across all their data, testing custom AI models with autonomous lunar rovers, and expanding agentic workflows across New Glenn orbital rockets and lunar lander programs.

“With 95% of software engineers already using generative AI tools to write code, agents accelerating approvals, and manufacturing teams resolving issues 70% faster, Blue Origin is proving that democratized AI adoption can both improve productivity and unlock innovation at scale.” Brennan concluded. "With AWS and agentic AI, we're building the road to space faster than ever before — for the benefit of Earth."