Retiring the “Human API”: Building the Agentic Future of Engineering
How AI can automate the future of engineering
Across manufacturing, robotics, and industrial equipment, physical product teams now operate under software‑like time pressure. Yet the core workflow of a mechanical engineer has barely changed in twenty years. The mechanical engineering stack has solidified into a fortress of systems of record: massive, expensive databases that are excellent at storing parts, assemblies, and bills of materials but blind to how AI could transform the actual work of engineering.
This growing mismatch between market expectations and operational reality has brought the industry to a breaking point. It’s time for a new approach.
The “Human API”
The entire tech world is currently fixated on OpenClaw, the open‑source autonomous agent that’s rapidly becoming a standard for practical automation. In a matter of months, it has evolved from a passionate GitHub experiment into a core part of countless developer workflows, showcasing what true system‑level autonomy looks like. Meanwhile, highly paid mechanical engineers are still forced to act as “human APIs” between disconnected manufacturing systems. They tab‑switch between SolidWorks, Excel, Jira, PLM, and ERP, spending roughly 30–40% of their time on non‑engineering work like updating BOMs, chasing vendor quotes, and manually checking compliance. The industry has poured investment into “digital thread” initiatives to connect CAD, PLM, and ERP, but these integrations are often brittle and fragile in real‑world use. We don’t need to replace senior experts; we need to codify their judgment into reusable patterns so that routine, repetitive tasks no longer consume their days.
The Rise of Physical AI
We are in the midst of a shift to Physical AI: intelligence that directly interacts with and optimizes the mechanical world. Automotive teams are continuously simulating structural behavior in the background, aerospace engineers are rapidly iterating geometries before physical testing, and electronics developers are increasingly blurring the lines between EDA and PLM. The direction of travel is clear: more of the engineering lifecycle is becoming computational, continuous, and data‑driven.
Yet the tools serving these engineers have not kept pace. Legacy CAD and PLM incumbents are constrained by their business models, bolting on basic chatbots and incremental assistants rather than rethinking the underlying workflow. At the same time, early startups have introduced capabilities like “text‑to‑CAD” but often treat them as isolated novelties rather than integral components of a robust engineering stack. Generative geometry is powerful, but you cannot simply drop a generic chatbot into an assembly‑line environment and expect it to function within strict production, safety, and quality constraints.
Building the Agentic Future
The future of engineering won’t be driven by issuing low‑level commands to databases. Unlocking real value from foundation models in mechanical engineering requires a new layer: agentic workflows. These are systems of action, not just systems of record, where software agents reason over goals, constraints, and live operational data, coordinate among themselves, and escalate only the hard tradeoffs to humans.
In this world, senior experts remain at the center, but their judgment is captured, structured, and re‑applied at scale. Agents encode patterns such as manufacturability rules, organizational standards, and supplier preferences, and they continuously enforce them across tools. Instead of engineers orchestrating every step between CAD, PLM, ERP, and the factory floor, agents handle the coordination, freeing humans to focus on genuinely hard problems in design, safety, and performance.
We are building this future.
If building the agentic future of mechanical engineering resonates with you, especially if you’ve felt the pain of acting as the “human API” between CAD, PLM, ERP, and the factory, we’d love to hear from you. Learn more and view open roles here.