Agentic AI for Engineering: From Design to Maintenance

June 2025

Organizations building and maintaining engineering systems are bottlenecked by human speed - The process of building and maintaining physical engineering systems, from bicycles to Apollo 11, has remained more or less the same over the last century. Designing these systems requires human labor that is expensive, hard to train/replace and varying in motivation/competence. The learning cycle of designing, implementing, validating and maintaining these systems is also limited by human labor. 

So how does this process play out today?  You start with the question of what you want to build. System engineers write down the system-level performance and design requirements. Junior engineers, working at the decomposed component level, define the architecture within their scope, and perform Computer Aided Design/Engineering (CAD/CAE) simulations to design the components. Manufacturing teams validate that designs are manufacturable and build them in collaboration with suppliers. The system is then deployed in the field and customer support engineers review the system logs, perform maintenance, identify the root cause in case of issues, and ensure the system performs as it should. 

Engineering companies are typically organized to mirror this process from research to design to manufacturing and support. Depending on system complexity, thousands of engineers are constantly working within their scope while ensuring they communicate across functions so the system runs as required.

The opportunity: AI Agents across the lifecycle - AI agents like Claude Code/Gemini CLI/OpenAI Codex and coding assistants like Cursor/Github copilot are already accelerating software development. But the opportunity in physical engineering is even greater. 

Why? Because current engineering workflows are bottlenecked by CAD tool complexity, fragmented documents and tribal knowledge. AI agents can act as copilots to unblock engineers across the lifecycle of physical engineering systems. Here are some obvious examples: 

1. In the requirements phase, AI agents can help engineers write and revise requirement documents, based on established ways of working. 
2. In the design & implementation phase, AI agents can search CAD/CAE tool documentation, retrieve relevant configurations, and eventually augment expensive simulation workflows with surrogate models (AI enabled digital twins).  
3. In the manufacturing phase, AI agents can help generate step-by-step instructions for specific tasks on the factory floor. 
4. In the maintenance phase, AI agents can parse system logs, correlate with past failures, and suggest diagnostics to help identify and fix root cause errors. 

Much of engineering time isn’t lost in invention, but in finding and using the right tool, test, config, or document. That’s exactly where AI agents will shine. They will start contributing by removing information and search bottlenecks where engineering time is most often wasted. 

Start Small: Intern-Level AI Agents First - Agentic AI for physical engineering won’t begin by disrupting org charts. It will still start with having intern level copilots that complement human engineers on low level tasks like finding a test script or summarizing a log file. As the AI agents become better at assisting with low level tasks and earn the confidence of human engineers, they will get pushed up the value chain and be trusted to operate more autonomously. 

How else will things change for engineering organizations?  AI will also disrupt generic operations of engineering organizations, such as IT, legal, and HR. But these functions are not unique to engineering organizations building physical systems. Hyperscalers (e.g. Amazon, Google, Microsoft) and startups in verticals (e.g. Harvey in legal) are already well equipped to attack these operational use cases.

The real opportunity lies in engineering-native workflows, CAD/CAE, procurement, test benches, logs. This is where general purpose copilots will fall short and purpose-built AI agents for engineering will be needed. 

Why this disruption matters - Organizations that successfully integrate Agentic AI in their process of engineering design and operations will change the pace at which they can learn and build products. Faster iteration loops mean tighter integration between design, manufacturing, deployment, and field feedback. Engineering organizations armed with AI agents will outcompete legacy competitors, compounding product value over time. 

Where to start: Customer Support - Complexity increases as you move upstream from customer support, to manufacturing, to design and eventually to research. The earlier phases are more ambiguous and open-ended, while later stages like customer support are narrower in scope. 

That’s why we may want to begin at the end, i.e. customer support, which is a high friction, but immediate value use case. Today, root cause identification is manual, engineers lose hours digging through past issues. The customer support problem for engineering systems has a clear scope, the feedback loops are fast, and the success criteria is clear. AI agents can assist immediately, prove value and then climb up the lifecycle.

Take me home