Intelligence: AI in Robotics

The Software Obsession

Everyone in robotics talks about the software. The investors, the founders, the journalists. It is all AI, machine learning, foundation models. The pitch decks glow with neural network diagrams. The press releases announce breakthroughs in “physical AI.” The valuations reflect software-like expectations for companies that also have to build, ship, and maintain physical machines.

Here is the problem: most robotics companies that have failed in the past decade did not fail because their AI was bad. They failed because the robot could not survive the loading dock. Because the gripper jammed. Because field service costs ate the margin. Because the sales cycle for selling hardware into manufacturing is 18 months, not 18 days.