Key Takeaways

  • Arda partnered with defense contractor Anduril to cut virtual build times for mechanical actuators from days down to 90 minutes.
  • Manufacturing engineering lacks an equivalent to Git or an IDE, leaving factories without a single source of truth for build plans.
  • McGrew rejects humanoid robots for manufacturing, arguing that one or two standard arms with a camera solve closed-world factory automation.
  • Flexible assembly lines require software that can reconfigure physical tooling and spin production up or down as demand shifts.

The Missing IDE for Hardware

Software engineers take Git and their IDE for granted. When a developer writes code, version control tracks every file, the compiler catches syntax errors, and tools like Cursor suggest logic in real time. Physical manufacturing has never had that baseline. An engineer designing an assembly line works across fragmented CAD drawings, static spreadsheets, and tribal shop-floor knowledge.

Bob McGrew saw this gap while starting Arda. After serving as research chief at OpenAI and running teams at Palantir, McGrew chose to focus on factory operations. He realized that before AI agents can automate a factory floor, manufacturing needs a digital foundation. As McGrew put it: “We are lucky that we can start by saying, 'Okay, we wanna be Cursor here,' effectively. Noting that Cursor basically said, 'Okay, we're gonna take this existing IDE, and we're gonna add an agent.' There's no IDE for manufacturing engineers. So first we had to build that so that there's a true data structure, there's the equivalent of, of effectively Git that you can work with.” (play from 28:10)

Arda spent the past year building this platform alongside Anduril. The defense contractor needed a single source of truth to coordinate assembly plans across complex defense systems, including combat aircraft like the Fury. McGrew noted that Anduril leaders “talked about the need for a single source of truth for manufacturing engineering, where one place where you actually encapsulated the manufacturing process and the manufacturing plan... and a reconfigurable factory that would allow you to change the manufacturing process and then also spin up and spin down different lines based on demand.” (play from 26:28)

The productivity gains showed up immediately on specific components. An Anduril engineer working on an actuator needed to plan out the virtual assembly sequence. “Normally that would've taken him a few days in order to do the virtual build for this,” (play from 27:04) McGrew explained, “and he was able to do that in 90 minutes using our software.” (play from 27:08)

Closed-World Robotics Beats Humanoids

Much of the venture capital in robotics chases humanoids designed to walk around unstructured, open worlds. McGrew thinks that strategy is backward for manufacturing. A factory is not an open street or a messy living room. It is a structured, closed environment where you control the lighting, the parts presentation, and the workstation bounds.

Because the operational envelope is closed, you do not need bipedal robots or multi-million-dollar custom machines. You need simple mechanical arms guided by machine vision and modern motion planning. McGrew was blunt about the hardware requirements: “We don't need to work in automotive where they have these, like, extremely large baroque robotics. I think, you know, two arms and a camera is enough. In a lot of cases, actually one arm and a camera is enough, and maybe you need some more, some better sensing.” (play from 20:04)

The lesson for technical founders is clear. If you want to automate physical work, define the tightest possible envelope. When you treat the assembly cell as a closed box, the hardware problem collapses into a routine kinematic task. The real challenge shifts to the software: tracking parts state, planning collision-free motion paths, and letting manufacturing engineers update the build sequence in software before metal touches metal.

What to Do With This

Audit your operational stack for your single source of truth. If your engineers use spreadsheets to bridge data between separate CAD and shop floor tools, stop writing custom glue scripts. Pick your highest-friction subassembly, model its full assembly steps inside a unified schema, and measure whether you can reconfigure the line without manual shop-floor retooling.